diff --git a/0321_market_vit_base/test_log.txt b/0321_market_vit_base/test_log.txt new file mode 100644 index 000000000..16a7a8b7e --- /dev/null +++ b/0321_market_vit_base/test_log.txt @@ -0,0 +1,144 @@ +2022-11-05 10:09:15,520 transreid INFO: Namespace(config_file='D:\\\\TransReID\\\\configs\\\\Market\\\\swin.yml', opts=['MODEL.DEVICE_ID', "('0')"]) +2022-11-05 10:09:15,520 transreid INFO: Loaded configuration file D:\\TransReID\\configs\\Market\\swin.yml +2022-11-05 10:09:15,520 transreid INFO: +MODEL: + PRETRAIN_CHOICE: 'imagenet' + PRETRAIN_PATH: 'D:\\TransReID\\swin_tiny_patch4_window7_224.pth' + METRIC_LOSS_TYPE: 'triplet' + IF_LABELSMOOTH: 'off' + IF_WITH_CENTER: 'no' + NAME: 'transformer' + NO_MARGIN: True + DEVICE_ID: ('0') + TRANSFORMER_TYPE: 'swin' + STRIDE_SIZE: [16, 16] + +INPUT: + SIZE_TRAIN: [256, 128] + SIZE_TEST: [256, 128] + PROB: 0.5 # random horizontal flip + RE_PROB: 0.5 # random erasing + PADDING: 10 + PIXEL_MEAN: [0.5, 0.5, 0.5] + PIXEL_STD: [0.5, 0.5, 0.5] + +DATASETS: + NAMES: ('market1501') + ROOT_DIR: ('D:\\TransReID') + +DATALOADER: + SAMPLER: 'softmax_triplet' + NUM_INSTANCE: 4 + NUM_WORKERS: 0 + +SOLVER: + OPTIMIZER_NAME: 'SGD' + MAX_EPOCHS: 120 + BASE_LR: 0.008 + IMS_PER_BATCH: 64 + WARMUP_METHOD: 'linear' + LARGE_FC_LR: False + CHECKPOINT_PERIOD: 120 + LOG_PERIOD: 50 + EVAL_PERIOD: 120 + WEIGHT_DECAY: 1e-4 + WEIGHT_DECAY_BIAS: 1e-4 + BIAS_LR_FACTOR: 2 + +TEST: + EVAL: True + IMS_PER_BATCH: 256 + RE_RANKING: False + WEIGHT: 'D:\\TransReID\\0321_market_vit_base\\transformer_120.pth' + NECK_FEAT: 'before' + FEAT_NORM: 'yes' + +OUTPUT_DIR: 'D:\\TransReID\\0321_market_vit_base' + + + +2022-11-05 10:09:15,520 transreid INFO: Running with config: +DATALOADER: + NUM_INSTANCE: 4 + NUM_WORKERS: 0 + SAMPLER: softmax_triplet +DATASETS: + NAMES: market1501 + ROOT_DIR: D:\TransReID +INPUT: + PADDING: 10 + PIXEL_MEAN: [0.5, 0.5, 0.5] + PIXEL_STD: [0.5, 0.5, 0.5] + PROB: 0.5 + RE_PROB: 0.5 + SIZE_TEST: [256, 128] + SIZE_TRAIN: [256, 128] +MODEL: + ATT_DROP_RATE: 0.0 + COS_LAYER: False + DEVICE: cuda + DEVICE_ID: 0 + DEVIDE_LENGTH: 4 + DIST_TRAIN: False + DROP_OUT: 0.0 + DROP_PATH: 0.1 + ID_LOSS_TYPE: softmax + ID_LOSS_WEIGHT: 1.0 + IF_LABELSMOOTH: off + IF_WITH_CENTER: no + JPM: False + LAST_STRIDE: 1 + METRIC_LOSS_TYPE: triplet + NAME: transformer + NECK: bnneck + NO_MARGIN: True + PRETRAIN_CHOICE: imagenet + PRETRAIN_PATH: D:\\TransReID\\swin_tiny_patch4_window7_224.pth + RE_ARRANGE: True + SHIFT_NUM: 5 + SHUFFLE_GROUP: 2 + SIE_CAMERA: False + SIE_COE: 3.0 + SIE_VIEW: False + STRIDE_SIZE: [16, 16] + TRANSFORMER_TYPE: swin + TRIPLET_LOSS_WEIGHT: 1.0 +OUTPUT_DIR: D:\\TransReID\\0321_market_vit_base +SOLVER: + BASE_LR: 0.008 + BIAS_LR_FACTOR: 2 + CENTER_LOSS_WEIGHT: 0.0005 + CENTER_LR: 0.5 + CHECKPOINT_PERIOD: 120 + COSINE_MARGIN: 0.5 + COSINE_SCALE: 30 + EVAL_PERIOD: 120 + GAMMA: 0.1 + IMS_PER_BATCH: 64 + LARGE_FC_LR: False + LOG_PERIOD: 50 + MARGIN: 0.3 + MAX_EPOCHS: 120 + MOMENTUM: 0.9 + OPTIMIZER_NAME: SGD + SEED: 1234 + STEPS: (40, 70) + WARMUP_EPOCHS: 5 + WARMUP_FACTOR: 0.01 + WARMUP_METHOD: linear + WEIGHT_DECAY: 0.0001 + WEIGHT_DECAY_BIAS: 0.0001 +TEST: + DIST_MAT: dist_mat.npy + EVAL: True + FEAT_NORM: yes + IMS_PER_BATCH: 256 + NECK_FEAT: before + RE_RANKING: False + WEIGHT: D:\\TransReID\\0321_market_vit_base\\transformer_120.pth +2022-11-05 10:09:16,585 transreid.test INFO: Enter inferencing +2022-11-05 10:10:41,395 transreid.test INFO: Validation Results +2022-11-05 10:10:41,396 transreid.test INFO: mAP: 81.9% +2022-11-05 10:10:41,396 transreid.test INFO: CMC curve, Rank-1 :92.8% +2022-11-05 10:10:41,396 transreid.test INFO: CMC curve, Rank-5 :97.5% +2022-11-05 10:10:41,396 transreid.test INFO: CMC curve, Rank-10 :98.4% diff --git a/0321_market_vit_base/train_log.txt b/0321_market_vit_base/train_log.txt new file mode 100644 index 000000000..27356441e --- /dev/null +++ b/0321_market_vit_base/train_log.txt @@ -0,0 +1,625 @@ +2022-11-04 21:06:52,939 transreid INFO: Saving model in the path :D:\\TransReID\\0321_market_vit_base +2022-11-04 21:06:52,939 transreid INFO: Namespace(config_file='D:\\\\TransReID\\\\configs\\\\Market\\\\swin.yml', local_rank=0, opts=['MODEL.DEVICE_ID', "('0')"]) +2022-11-04 21:06:52,940 transreid INFO: Loaded configuration file D:\\TransReID\\configs\\Market\\swin.yml +2022-11-04 21:06:52,940 transreid INFO: +MODEL: + PRETRAIN_CHOICE: 'imagenet' + PRETRAIN_PATH: 'D:\\TransReID\\swin_tiny_patch4_window7_224.pth' + METRIC_LOSS_TYPE: 'triplet' + IF_LABELSMOOTH: 'off' + IF_WITH_CENTER: 'no' + NAME: 'transformer' + NO_MARGIN: True + DEVICE_ID: ('0') + TRANSFORMER_TYPE: 'swin' + STRIDE_SIZE: [16, 16] + +INPUT: + SIZE_TRAIN: [256, 128] + SIZE_TEST: [256, 128] + PROB: 0.5 # random horizontal flip + RE_PROB: 0.5 # random erasing + PADDING: 10 + PIXEL_MEAN: [0.5, 0.5, 0.5] + PIXEL_STD: [0.5, 0.5, 0.5] + +DATASETS: + NAMES: ('market1501') + ROOT_DIR: ('D:\\TransReID') + +DATALOADER: + SAMPLER: 'softmax_triplet' + NUM_INSTANCE: 4 + NUM_WORKERS: 0 + +SOLVER: + OPTIMIZER_NAME: 'SGD' + MAX_EPOCHS: 120 + BASE_LR: 0.008 + IMS_PER_BATCH: 64 + WARMUP_METHOD: 'linear' + LARGE_FC_LR: False + CHECKPOINT_PERIOD: 120 + LOG_PERIOD: 50 + EVAL_PERIOD: 120 + WEIGHT_DECAY: 1e-4 + WEIGHT_DECAY_BIAS: 1e-4 + BIAS_LR_FACTOR: 2 + +TEST: + EVAL: True + IMS_PER_BATCH: 256 + RE_RANKING: False + WEIGHT: 'D:\\TransReID\\0321_market_vit_base\\swin.pth' + NECK_FEAT: 'before' + FEAT_NORM: 'yes' + +OUTPUT_DIR: 'D:\\TransReID\\0321_market_vit_base' + + + +2022-11-04 21:06:52,940 transreid INFO: Running with config: +DATALOADER: + NUM_INSTANCE: 4 + NUM_WORKERS: 0 + SAMPLER: softmax_triplet +DATASETS: + NAMES: market1501 + ROOT_DIR: D:\TransReID +INPUT: + PADDING: 10 + PIXEL_MEAN: [0.5, 0.5, 0.5] + PIXEL_STD: [0.5, 0.5, 0.5] + PROB: 0.5 + RE_PROB: 0.5 + SIZE_TEST: [256, 128] + SIZE_TRAIN: [256, 128] +MODEL: + ATT_DROP_RATE: 0.0 + COS_LAYER: False + DEVICE: cuda + DEVICE_ID: 0 + DEVIDE_LENGTH: 4 + DIST_TRAIN: False + DROP_OUT: 0.0 + DROP_PATH: 0.1 + ID_LOSS_TYPE: softmax + ID_LOSS_WEIGHT: 1.0 + IF_LABELSMOOTH: off + IF_WITH_CENTER: no + JPM: False + LAST_STRIDE: 1 + METRIC_LOSS_TYPE: triplet + NAME: transformer + NECK: bnneck + NO_MARGIN: True + PRETRAIN_CHOICE: imagenet + PRETRAIN_PATH: D:\\TransReID\\swin_tiny_patch4_window7_224.pth + RE_ARRANGE: True + SHIFT_NUM: 5 + SHUFFLE_GROUP: 2 + SIE_CAMERA: False + SIE_COE: 3.0 + SIE_VIEW: False + STRIDE_SIZE: [16, 16] + TRANSFORMER_TYPE: swin + TRIPLET_LOSS_WEIGHT: 1.0 +OUTPUT_DIR: D:\\TransReID\\0321_market_vit_base +SOLVER: + BASE_LR: 0.008 + BIAS_LR_FACTOR: 2 + CENTER_LOSS_WEIGHT: 0.0005 + CENTER_LR: 0.5 + CHECKPOINT_PERIOD: 120 + COSINE_MARGIN: 0.5 + COSINE_SCALE: 30 + EVAL_PERIOD: 120 + GAMMA: 0.1 + IMS_PER_BATCH: 64 + LARGE_FC_LR: False + LOG_PERIOD: 50 + MARGIN: 0.3 + MAX_EPOCHS: 120 + MOMENTUM: 0.9 + OPTIMIZER_NAME: SGD + SEED: 1234 + STEPS: (40, 70) + WARMUP_EPOCHS: 5 + WARMUP_FACTOR: 0.01 + WARMUP_METHOD: linear + WEIGHT_DECAY: 0.0001 + WEIGHT_DECAY_BIAS: 0.0001 +TEST: + DIST_MAT: dist_mat.npy + EVAL: True + FEAT_NORM: yes + IMS_PER_BATCH: 256 + NECK_FEAT: before + RE_RANKING: False + WEIGHT: D:\\TransReID\\0321_market_vit_base\\swin.pth +2022-11-04 21:06:53,596 transreid.train INFO: start training +2022-11-04 21:07:16,803 transreid.train INFO: Epoch[1] Iteration[50/186] Loss: 7.901, Acc: 0.010, Base Lr: 1.66e-03 +2022-11-04 21:07:36,527 transreid.train INFO: Epoch[1] Iteration[100/186] Loss: 7.649, Acc: 0.024, Base Lr: 1.66e-03 +2022-11-04 21:07:56,339 transreid.train INFO: Epoch[1] Iteration[150/186] Loss: 7.476, Acc: 0.029, Base Lr: 1.66e-03 +2022-11-04 21:08:09,433 transreid.train INFO: Epoch 1 done. Time per batch: 0.414[s] Speed: 154.5[samples/s] +2022-11-04 21:08:28,589 transreid.train INFO: Epoch[2] Iteration[50/186] Loss: 6.657, Acc: 0.047, Base Lr: 3.25e-03 +2022-11-04 21:08:47,666 transreid.train INFO: Epoch[2] Iteration[100/186] Loss: 6.387, Acc: 0.065, Base Lr: 3.25e-03 +2022-11-04 21:09:06,741 transreid.train INFO: Epoch[2] Iteration[150/186] Loss: 6.097, Acc: 0.096, Base Lr: 3.25e-03 +2022-11-04 21:09:19,752 transreid.train INFO: Epoch 2 done. Time per batch: 0.382[s] Speed: 167.5[samples/s] +2022-11-04 21:09:38,862 transreid.train INFO: Epoch[3] Iteration[50/186] Loss: 5.362, Acc: 0.145, Base Lr: 4.83e-03 +2022-11-04 21:09:58,015 transreid.train INFO: Epoch[3] Iteration[100/186] Loss: 4.979, Acc: 0.213, Base Lr: 4.83e-03 +2022-11-04 21:10:17,190 transreid.train INFO: Epoch[3] Iteration[150/186] Loss: 4.586, Acc: 0.266, Base Lr: 4.83e-03 +2022-11-04 21:10:29,938 transreid.train INFO: Epoch 3 done. Time per batch: 0.384[s] Speed: 166.9[samples/s] +2022-11-04 21:10:49,788 transreid.train INFO: Epoch[4] Iteration[50/186] Loss: 4.075, Acc: 0.315, Base Lr: 6.42e-03 +2022-11-04 21:11:09,106 transreid.train INFO: Epoch[4] Iteration[100/186] Loss: 3.695, Acc: 0.379, Base Lr: 6.42e-03 +2022-11-04 21:11:29,172 transreid.train INFO: Epoch[4] Iteration[150/186] Loss: 3.335, Acc: 0.448, Base Lr: 6.42e-03 +2022-11-04 21:11:42,505 transreid.train INFO: Epoch 4 done. Time per batch: 0.397[s] Speed: 161.4[samples/s] +2022-11-04 21:12:02,033 transreid.train INFO: Epoch[5] Iteration[50/186] Loss: 3.213, Acc: 0.460, Base Lr: 7.97e-03 +2022-11-04 21:12:21,490 transreid.train INFO: Epoch[5] Iteration[100/186] Loss: 2.794, Acc: 0.534, Base Lr: 7.97e-03 +2022-11-04 21:12:41,212 transreid.train INFO: Epoch[5] Iteration[150/186] Loss: 2.485, Acc: 0.595, Base Lr: 7.97e-03 +2022-11-04 21:12:54,221 transreid.train INFO: Epoch 5 done. Time per batch: 0.392[s] Speed: 163.3[samples/s] +2022-11-04 21:13:14,412 transreid.train INFO: Epoch[6] Iteration[50/186] Loss: 2.213, Acc: 0.619, Base Lr: 7.95e-03 +2022-11-04 21:13:34,192 transreid.train INFO: Epoch[6] Iteration[100/186] Loss: 1.961, Acc: 0.675, Base Lr: 7.95e-03 +2022-11-04 21:13:53,394 transreid.train INFO: Epoch[6] Iteration[150/186] Loss: 1.796, Acc: 0.707, Base Lr: 7.95e-03 +2022-11-04 21:14:06,022 transreid.train INFO: Epoch 6 done. Time per batch: 0.392[s] Speed: 163.1[samples/s] +2022-11-04 21:14:25,172 transreid.train INFO: Epoch[7] Iteration[50/186] Loss: 1.750, Acc: 0.700, Base Lr: 7.93e-03 +2022-11-04 21:14:44,355 transreid.train INFO: Epoch[7] Iteration[100/186] Loss: 1.512, Acc: 0.757, Base Lr: 7.93e-03 +2022-11-04 21:15:03,520 transreid.train INFO: Epoch[7] Iteration[150/186] Loss: 1.336, Acc: 0.792, Base Lr: 7.93e-03 +2022-11-04 21:15:16,167 transreid.train INFO: Epoch 7 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:15:35,335 transreid.train INFO: Epoch[8] Iteration[50/186] Loss: 1.184, Acc: 0.801, Base Lr: 7.91e-03 +2022-11-04 21:15:54,499 transreid.train INFO: Epoch[8] Iteration[100/186] Loss: 1.048, Acc: 0.833, Base Lr: 7.91e-03 +2022-11-04 21:16:13,677 transreid.train INFO: Epoch[8] Iteration[150/186] Loss: 0.968, Acc: 0.855, Base Lr: 7.91e-03 +2022-11-04 21:16:26,353 transreid.train INFO: Epoch 8 done. Time per batch: 0.384[s] Speed: 166.9[samples/s] +2022-11-04 21:16:45,582 transreid.train INFO: Epoch[9] Iteration[50/186] Loss: 0.977, Acc: 0.855, Base Lr: 7.89e-03 +2022-11-04 21:17:04,783 transreid.train INFO: Epoch[9] Iteration[100/186] Loss: 0.907, Acc: 0.867, Base Lr: 7.89e-03 +2022-11-04 21:17:23,987 transreid.train INFO: Epoch[9] Iteration[150/186] Loss: 0.821, Acc: 0.887, Base Lr: 7.89e-03 +2022-11-04 21:17:36,648 transreid.train INFO: Epoch 9 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 21:17:55,842 transreid.train INFO: Epoch[10] Iteration[50/186] Loss: 0.866, Acc: 0.870, Base Lr: 7.86e-03 +2022-11-04 21:18:14,994 transreid.train INFO: Epoch[10] Iteration[100/186] Loss: 0.791, Acc: 0.889, Base Lr: 7.86e-03 +2022-11-04 21:18:34,187 transreid.train INFO: Epoch[10] Iteration[150/186] Loss: 0.694, Acc: 0.905, Base Lr: 7.86e-03 +2022-11-04 21:18:46,860 transreid.train INFO: Epoch 10 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:19:06,060 transreid.train INFO: Epoch[11] Iteration[50/186] Loss: 0.722, Acc: 0.902, Base Lr: 7.84e-03 +2022-11-04 21:19:25,246 transreid.train INFO: Epoch[11] Iteration[100/186] Loss: 0.659, Acc: 0.913, Base Lr: 7.84e-03 +2022-11-04 21:19:44,434 transreid.train INFO: Epoch[11] Iteration[150/186] Loss: 0.612, Acc: 0.923, Base Lr: 7.84e-03 +2022-11-04 21:19:57,451 transreid.train INFO: Epoch 11 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:20:16,635 transreid.train INFO: Epoch[12] Iteration[50/186] Loss: 0.650, Acc: 0.911, Base Lr: 7.80e-03 +2022-11-04 21:20:35,787 transreid.train INFO: Epoch[12] Iteration[100/186] Loss: 0.580, Acc: 0.923, Base Lr: 7.80e-03 +2022-11-04 21:20:54,930 transreid.train INFO: Epoch[12] Iteration[150/186] Loss: 0.533, Acc: 0.933, Base Lr: 7.80e-03 +2022-11-04 21:21:07,967 transreid.train INFO: Epoch 12 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:21:27,122 transreid.train INFO: Epoch[13] Iteration[50/186] Loss: 0.525, Acc: 0.927, Base Lr: 7.77e-03 +2022-11-04 21:21:46,290 transreid.train INFO: Epoch[13] Iteration[100/186] Loss: 0.485, Acc: 0.936, Base Lr: 7.77e-03 +2022-11-04 21:22:05,464 transreid.train INFO: Epoch[13] Iteration[150/186] Loss: 0.461, Acc: 0.942, Base Lr: 7.77e-03 +2022-11-04 21:22:18,524 transreid.train INFO: Epoch 13 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 21:22:37,722 transreid.train INFO: Epoch[14] Iteration[50/186] Loss: 0.468, Acc: 0.936, Base Lr: 7.73e-03 +2022-11-04 21:22:56,877 transreid.train INFO: Epoch[14] Iteration[100/186] Loss: 0.431, Acc: 0.946, Base Lr: 7.73e-03 +2022-11-04 21:23:16,047 transreid.train INFO: Epoch[14] Iteration[150/186] Loss: 0.405, Acc: 0.947, Base Lr: 7.73e-03 +2022-11-04 21:23:29,090 transreid.train INFO: Epoch 14 done. Time per batch: 0.384[s] Speed: 166.9[samples/s] +2022-11-04 21:23:48,278 transreid.train INFO: Epoch[15] Iteration[50/186] Loss: 0.431, Acc: 0.949, Base Lr: 7.70e-03 +2022-11-04 21:24:07,439 transreid.train INFO: Epoch[15] Iteration[100/186] Loss: 0.389, Acc: 0.950, Base Lr: 7.70e-03 +2022-11-04 21:24:26,576 transreid.train INFO: Epoch[15] Iteration[150/186] Loss: 0.361, Acc: 0.955, Base Lr: 7.70e-03 +2022-11-04 21:24:39,237 transreid.train INFO: Epoch 15 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:24:58,423 transreid.train INFO: Epoch[16] Iteration[50/186] Loss: 0.341, Acc: 0.955, Base Lr: 7.65e-03 +2022-11-04 21:25:17,570 transreid.train INFO: Epoch[16] Iteration[100/186] Loss: 0.344, Acc: 0.957, Base Lr: 7.65e-03 +2022-11-04 21:25:36,710 transreid.train INFO: Epoch[16] Iteration[150/186] Loss: 0.328, Acc: 0.961, Base Lr: 7.65e-03 +2022-11-04 21:25:49,755 transreid.train INFO: Epoch 16 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:26:08,964 transreid.train INFO: Epoch[17] Iteration[50/186] Loss: 0.356, Acc: 0.958, Base Lr: 7.61e-03 +2022-11-04 21:26:28,125 transreid.train INFO: Epoch[17] Iteration[100/186] Loss: 0.334, Acc: 0.960, Base Lr: 7.61e-03 +2022-11-04 21:26:47,317 transreid.train INFO: Epoch[17] Iteration[150/186] Loss: 0.307, Acc: 0.964, Base Lr: 7.61e-03 +2022-11-04 21:26:59,976 transreid.train INFO: Epoch 17 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:27:19,153 transreid.train INFO: Epoch[18] Iteration[50/186] Loss: 0.304, Acc: 0.961, Base Lr: 7.56e-03 +2022-11-04 21:27:38,292 transreid.train INFO: Epoch[18] Iteration[100/186] Loss: 0.327, Acc: 0.960, Base Lr: 7.56e-03 +2022-11-04 21:27:57,439 transreid.train INFO: Epoch[18] Iteration[150/186] Loss: 0.307, Acc: 0.964, Base Lr: 7.56e-03 +2022-11-04 21:28:10,497 transreid.train INFO: Epoch 18 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:28:29,656 transreid.train INFO: Epoch[19] Iteration[50/186] Loss: 0.281, Acc: 0.967, Base Lr: 7.52e-03 +2022-11-04 21:28:48,847 transreid.train INFO: Epoch[19] Iteration[100/186] Loss: 0.269, Acc: 0.968, Base Lr: 7.52e-03 +2022-11-04 21:29:08,041 transreid.train INFO: Epoch[19] Iteration[150/186] Loss: 0.260, Acc: 0.971, Base Lr: 7.52e-03 +2022-11-04 21:29:20,703 transreid.train INFO: Epoch 19 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:29:39,863 transreid.train INFO: Epoch[20] Iteration[50/186] Loss: 0.280, Acc: 0.964, Base Lr: 7.47e-03 +2022-11-04 21:29:59,010 transreid.train INFO: Epoch[20] Iteration[100/186] Loss: 0.260, Acc: 0.968, Base Lr: 7.47e-03 +2022-11-04 21:30:18,200 transreid.train INFO: Epoch[20] Iteration[150/186] Loss: 0.249, Acc: 0.971, Base Lr: 7.47e-03 +2022-11-04 21:30:30,834 transreid.train INFO: Epoch 20 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:30:50,047 transreid.train INFO: Epoch[21] Iteration[50/186] Loss: 0.277, Acc: 0.965, Base Lr: 7.41e-03 +2022-11-04 21:31:09,186 transreid.train INFO: Epoch[21] Iteration[100/186] Loss: 0.269, Acc: 0.970, Base Lr: 7.41e-03 +2022-11-04 21:31:28,323 transreid.train INFO: Epoch[21] Iteration[150/186] Loss: 0.240, Acc: 0.973, Base Lr: 7.41e-03 +2022-11-04 21:31:40,936 transreid.train INFO: Epoch 21 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 21:32:00,139 transreid.train INFO: Epoch[22] Iteration[50/186] Loss: 0.241, Acc: 0.975, Base Lr: 7.36e-03 +2022-11-04 21:32:19,286 transreid.train INFO: Epoch[22] Iteration[100/186] Loss: 0.226, Acc: 0.977, Base Lr: 7.36e-03 +2022-11-04 21:32:38,438 transreid.train INFO: Epoch[22] Iteration[150/186] Loss: 0.220, Acc: 0.978, Base Lr: 7.36e-03 +2022-11-04 21:32:51,106 transreid.train INFO: Epoch 22 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 21:33:10,286 transreid.train INFO: Epoch[23] Iteration[50/186] Loss: 0.218, Acc: 0.975, Base Lr: 7.30e-03 +2022-11-04 21:33:29,441 transreid.train INFO: Epoch[23] Iteration[100/186] Loss: 0.219, Acc: 0.975, Base Lr: 7.30e-03 +2022-11-04 21:33:48,644 transreid.train INFO: Epoch[23] Iteration[150/186] Loss: 0.214, Acc: 0.976, Base Lr: 7.30e-03 +2022-11-04 21:34:01,695 transreid.train INFO: Epoch 23 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:34:20,887 transreid.train INFO: Epoch[24] Iteration[50/186] Loss: 0.210, Acc: 0.977, Base Lr: 7.24e-03 +2022-11-04 21:34:40,047 transreid.train INFO: Epoch[24] Iteration[100/186] Loss: 0.199, Acc: 0.979, Base Lr: 7.24e-03 +2022-11-04 21:34:59,198 transreid.train INFO: Epoch[24] Iteration[150/186] Loss: 0.192, Acc: 0.981, Base Lr: 7.24e-03 +2022-11-04 21:35:11,840 transreid.train INFO: Epoch 24 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:35:31,012 transreid.train INFO: Epoch[25] Iteration[50/186] Loss: 0.175, Acc: 0.980, Base Lr: 7.18e-03 +2022-11-04 21:35:50,180 transreid.train INFO: Epoch[25] Iteration[100/186] Loss: 0.188, Acc: 0.981, Base Lr: 7.18e-03 +2022-11-04 21:36:09,324 transreid.train INFO: Epoch[25] Iteration[150/186] Loss: 0.180, Acc: 0.981, Base Lr: 7.18e-03 +2022-11-04 21:36:21,977 transreid.train INFO: Epoch 25 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:36:41,151 transreid.train INFO: Epoch[26] Iteration[50/186] Loss: 0.179, Acc: 0.977, Base Lr: 7.11e-03 +2022-11-04 21:37:00,279 transreid.train INFO: Epoch[26] Iteration[100/186] Loss: 0.174, Acc: 0.979, Base Lr: 7.11e-03 +2022-11-04 21:37:19,397 transreid.train INFO: Epoch[26] Iteration[150/186] Loss: 0.163, Acc: 0.980, Base Lr: 7.11e-03 +2022-11-04 21:37:32,040 transreid.train INFO: Epoch 26 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 21:37:51,246 transreid.train INFO: Epoch[27] Iteration[50/186] Loss: 0.188, Acc: 0.977, Base Lr: 7.04e-03 +2022-11-04 21:38:10,449 transreid.train INFO: Epoch[27] Iteration[100/186] Loss: 0.175, Acc: 0.980, Base Lr: 7.04e-03 +2022-11-04 21:38:29,583 transreid.train INFO: Epoch[27] Iteration[150/186] Loss: 0.168, Acc: 0.981, Base Lr: 7.04e-03 +2022-11-04 21:38:42,604 transreid.train INFO: Epoch 27 done. Time per batch: 0.384[s] Speed: 166.9[samples/s] +2022-11-04 21:39:01,811 transreid.train INFO: Epoch[28] Iteration[50/186] Loss: 0.144, Acc: 0.986, Base Lr: 6.97e-03 +2022-11-04 21:39:21,011 transreid.train INFO: Epoch[28] Iteration[100/186] Loss: 0.148, Acc: 0.984, Base Lr: 6.97e-03 +2022-11-04 21:39:40,187 transreid.train INFO: Epoch[28] Iteration[150/186] Loss: 0.154, Acc: 0.984, Base Lr: 6.97e-03 +2022-11-04 21:39:52,825 transreid.train INFO: Epoch 28 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:40:11,997 transreid.train INFO: Epoch[29] Iteration[50/186] Loss: 0.155, Acc: 0.980, Base Lr: 6.90e-03 +2022-11-04 21:40:31,148 transreid.train INFO: Epoch[29] Iteration[100/186] Loss: 0.134, Acc: 0.985, Base Lr: 6.90e-03 +2022-11-04 21:40:50,301 transreid.train INFO: Epoch[29] Iteration[150/186] Loss: 0.133, Acc: 0.986, Base Lr: 6.90e-03 +2022-11-04 21:41:02,944 transreid.train INFO: Epoch 29 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:41:22,113 transreid.train INFO: Epoch[30] Iteration[50/186] Loss: 0.142, Acc: 0.986, Base Lr: 6.83e-03 +2022-11-04 21:41:41,248 transreid.train INFO: Epoch[30] Iteration[100/186] Loss: 0.142, Acc: 0.986, Base Lr: 6.83e-03 +2022-11-04 21:42:00,436 transreid.train INFO: Epoch[30] Iteration[150/186] Loss: 0.136, Acc: 0.986, Base Lr: 6.83e-03 +2022-11-04 21:42:13,086 transreid.train INFO: Epoch 30 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:42:32,270 transreid.train INFO: Epoch[31] Iteration[50/186] Loss: 0.162, Acc: 0.986, Base Lr: 6.76e-03 +2022-11-04 21:42:51,448 transreid.train INFO: Epoch[31] Iteration[100/186] Loss: 0.152, Acc: 0.985, Base Lr: 6.76e-03 +2022-11-04 21:43:10,603 transreid.train INFO: Epoch[31] Iteration[150/186] Loss: 0.149, Acc: 0.986, Base Lr: 6.76e-03 +2022-11-04 21:43:23,609 transreid.train INFO: Epoch 31 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:43:42,811 transreid.train INFO: Epoch[32] Iteration[50/186] Loss: 0.167, Acc: 0.987, Base Lr: 6.68e-03 +2022-11-04 21:44:02,017 transreid.train INFO: Epoch[32] Iteration[100/186] Loss: 0.135, Acc: 0.988, Base Lr: 6.68e-03 +2022-11-04 21:44:21,188 transreid.train INFO: Epoch[32] Iteration[150/186] Loss: 0.128, Acc: 0.988, Base Lr: 6.68e-03 +2022-11-04 21:44:33,493 transreid.train INFO: Epoch 32 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 21:44:52,630 transreid.train INFO: Epoch[33] Iteration[50/186] Loss: 0.152, Acc: 0.986, Base Lr: 6.60e-03 +2022-11-04 21:45:11,822 transreid.train INFO: Epoch[33] Iteration[100/186] Loss: 0.152, Acc: 0.985, Base Lr: 6.60e-03 +2022-11-04 21:45:31,006 transreid.train INFO: Epoch[33] Iteration[150/186] Loss: 0.136, Acc: 0.987, Base Lr: 6.60e-03 +2022-11-04 21:45:43,715 transreid.train INFO: Epoch 33 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:46:02,897 transreid.train INFO: Epoch[34] Iteration[50/186] Loss: 0.125, Acc: 0.987, Base Lr: 6.52e-03 +2022-11-04 21:46:22,073 transreid.train INFO: Epoch[34] Iteration[100/186] Loss: 0.122, Acc: 0.988, Base Lr: 6.52e-03 +2022-11-04 21:46:41,271 transreid.train INFO: Epoch[34] Iteration[150/186] Loss: 0.127, Acc: 0.988, Base Lr: 6.52e-03 +2022-11-04 21:46:53,927 transreid.train INFO: Epoch 34 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:47:13,104 transreid.train INFO: Epoch[35] Iteration[50/186] Loss: 0.122, Acc: 0.987, Base Lr: 6.44e-03 +2022-11-04 21:47:32,277 transreid.train INFO: Epoch[35] Iteration[100/186] Loss: 0.134, Acc: 0.988, Base Lr: 6.44e-03 +2022-11-04 21:47:51,469 transreid.train INFO: Epoch[35] Iteration[150/186] Loss: 0.120, Acc: 0.989, Base Lr: 6.44e-03 +2022-11-04 21:48:04,513 transreid.train INFO: Epoch 35 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 21:48:23,693 transreid.train INFO: Epoch[36] Iteration[50/186] Loss: 0.141, Acc: 0.985, Base Lr: 6.35e-03 +2022-11-04 21:48:42,836 transreid.train INFO: Epoch[36] Iteration[100/186] Loss: 0.124, Acc: 0.987, Base Lr: 6.35e-03 +2022-11-04 21:49:02,008 transreid.train INFO: Epoch[36] Iteration[150/186] Loss: 0.118, Acc: 0.989, Base Lr: 6.35e-03 +2022-11-04 21:49:14,640 transreid.train INFO: Epoch 36 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:49:33,815 transreid.train INFO: Epoch[37] Iteration[50/186] Loss: 0.133, Acc: 0.989, Base Lr: 6.27e-03 +2022-11-04 21:49:52,951 transreid.train INFO: Epoch[37] Iteration[100/186] Loss: 0.118, Acc: 0.990, Base Lr: 6.27e-03 +2022-11-04 21:50:12,149 transreid.train INFO: Epoch[37] Iteration[150/186] Loss: 0.107, Acc: 0.990, Base Lr: 6.27e-03 +2022-11-04 21:50:24,766 transreid.train INFO: Epoch 37 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:50:43,955 transreid.train INFO: Epoch[38] Iteration[50/186] Loss: 0.099, Acc: 0.991, Base Lr: 6.18e-03 +2022-11-04 21:51:03,109 transreid.train INFO: Epoch[38] Iteration[100/186] Loss: 0.099, Acc: 0.993, Base Lr: 6.18e-03 +2022-11-04 21:51:22,254 transreid.train INFO: Epoch[38] Iteration[150/186] Loss: 0.098, Acc: 0.993, Base Lr: 6.18e-03 +2022-11-04 21:51:35,275 transreid.train INFO: Epoch 38 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 21:51:54,445 transreid.train INFO: Epoch[39] Iteration[50/186] Loss: 0.098, Acc: 0.989, Base Lr: 6.09e-03 +2022-11-04 21:52:13,573 transreid.train INFO: Epoch[39] Iteration[100/186] Loss: 0.097, Acc: 0.990, Base Lr: 6.09e-03 +2022-11-04 21:52:32,700 transreid.train INFO: Epoch[39] Iteration[150/186] Loss: 0.089, Acc: 0.991, Base Lr: 6.09e-03 +2022-11-04 21:52:45,731 transreid.train INFO: Epoch 39 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 21:53:04,917 transreid.train INFO: Epoch[40] Iteration[50/186] Loss: 0.121, Acc: 0.987, Base Lr: 6.00e-03 +2022-11-04 21:53:24,059 transreid.train INFO: Epoch[40] Iteration[100/186] Loss: 0.098, Acc: 0.990, Base Lr: 6.00e-03 +2022-11-04 21:53:43,194 transreid.train INFO: Epoch[40] Iteration[150/186] Loss: 0.097, Acc: 0.991, Base Lr: 6.00e-03 +2022-11-04 21:53:55,818 transreid.train INFO: Epoch 40 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 21:54:14,979 transreid.train INFO: Epoch[41] Iteration[50/186] Loss: 0.101, Acc: 0.989, Base Lr: 5.91e-03 +2022-11-04 21:54:34,096 transreid.train INFO: Epoch[41] Iteration[100/186] Loss: 0.095, Acc: 0.991, Base Lr: 5.91e-03 +2022-11-04 21:54:53,212 transreid.train INFO: Epoch[41] Iteration[150/186] Loss: 0.082, Acc: 0.993, Base Lr: 5.91e-03 +2022-11-04 21:55:05,830 transreid.train INFO: Epoch 41 done. Time per batch: 0.383[s] Speed: 167.3[samples/s] +2022-11-04 21:55:24,976 transreid.train INFO: Epoch[42] Iteration[50/186] Loss: 0.091, Acc: 0.991, Base Lr: 5.82e-03 +2022-11-04 21:55:44,143 transreid.train INFO: Epoch[42] Iteration[100/186] Loss: 0.102, Acc: 0.991, Base Lr: 5.82e-03 +2022-11-04 21:56:03,289 transreid.train INFO: Epoch[42] Iteration[150/186] Loss: 0.091, Acc: 0.992, Base Lr: 5.82e-03 +2022-11-04 21:56:15,933 transreid.train INFO: Epoch 42 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 21:56:35,065 transreid.train INFO: Epoch[43] Iteration[50/186] Loss: 0.074, Acc: 0.992, Base Lr: 5.73e-03 +2022-11-04 21:56:54,209 transreid.train INFO: Epoch[43] Iteration[100/186] Loss: 0.077, Acc: 0.992, Base Lr: 5.73e-03 +2022-11-04 21:57:13,346 transreid.train INFO: Epoch[43] Iteration[150/186] Loss: 0.077, Acc: 0.993, Base Lr: 5.73e-03 +2022-11-04 21:57:26,011 transreid.train INFO: Epoch 43 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 21:57:45,202 transreid.train INFO: Epoch[44] Iteration[50/186] Loss: 0.128, Acc: 0.989, Base Lr: 5.63e-03 +2022-11-04 21:58:04,367 transreid.train INFO: Epoch[44] Iteration[100/186] Loss: 0.110, Acc: 0.990, Base Lr: 5.63e-03 +2022-11-04 21:58:23,535 transreid.train INFO: Epoch[44] Iteration[150/186] Loss: 0.100, Acc: 0.990, Base Lr: 5.63e-03 +2022-11-04 21:58:36,584 transreid.train INFO: Epoch 44 done. Time per batch: 0.384[s] Speed: 166.9[samples/s] +2022-11-04 21:58:55,777 transreid.train INFO: Epoch[45] Iteration[50/186] Loss: 0.091, Acc: 0.992, Base Lr: 5.54e-03 +2022-11-04 21:59:14,938 transreid.train INFO: Epoch[45] Iteration[100/186] Loss: 0.083, Acc: 0.992, Base Lr: 5.54e-03 +2022-11-04 21:59:34,102 transreid.train INFO: Epoch[45] Iteration[150/186] Loss: 0.081, Acc: 0.992, Base Lr: 5.54e-03 +2022-11-04 21:59:46,732 transreid.train INFO: Epoch 45 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:00:05,931 transreid.train INFO: Epoch[46] Iteration[50/186] Loss: 0.113, Acc: 0.992, Base Lr: 5.44e-03 +2022-11-04 22:00:25,099 transreid.train INFO: Epoch[46] Iteration[100/186] Loss: 0.109, Acc: 0.992, Base Lr: 5.44e-03 +2022-11-04 22:00:44,265 transreid.train INFO: Epoch[46] Iteration[150/186] Loss: 0.097, Acc: 0.993, Base Lr: 5.44e-03 +2022-11-04 22:00:56,893 transreid.train INFO: Epoch 46 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:01:16,086 transreid.train INFO: Epoch[47] Iteration[50/186] Loss: 0.088, Acc: 0.992, Base Lr: 5.34e-03 +2022-11-04 22:01:35,258 transreid.train INFO: Epoch[47] Iteration[100/186] Loss: 0.073, Acc: 0.994, Base Lr: 5.34e-03 +2022-11-04 22:01:54,477 transreid.train INFO: Epoch[47] Iteration[150/186] Loss: 0.071, Acc: 0.995, Base Lr: 5.34e-03 +2022-11-04 22:02:07,149 transreid.train INFO: Epoch 47 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 22:02:26,369 transreid.train INFO: Epoch[48] Iteration[50/186] Loss: 0.095, Acc: 0.992, Base Lr: 5.24e-03 +2022-11-04 22:02:45,534 transreid.train INFO: Epoch[48] Iteration[100/186] Loss: 0.084, Acc: 0.994, Base Lr: 5.24e-03 +2022-11-04 22:03:04,706 transreid.train INFO: Epoch[48] Iteration[150/186] Loss: 0.081, Acc: 0.994, Base Lr: 5.24e-03 +2022-11-04 22:03:17,355 transreid.train INFO: Epoch 48 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 22:03:36,555 transreid.train INFO: Epoch[49] Iteration[50/186] Loss: 0.082, Acc: 0.992, Base Lr: 5.14e-03 +2022-11-04 22:03:55,713 transreid.train INFO: Epoch[49] Iteration[100/186] Loss: 0.079, Acc: 0.993, Base Lr: 5.14e-03 +2022-11-04 22:04:14,881 transreid.train INFO: Epoch[49] Iteration[150/186] Loss: 0.071, Acc: 0.993, Base Lr: 5.14e-03 +2022-11-04 22:04:27,507 transreid.train INFO: Epoch 49 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:04:46,657 transreid.train INFO: Epoch[50] Iteration[50/186] Loss: 0.075, Acc: 0.994, Base Lr: 5.04e-03 +2022-11-04 22:05:05,802 transreid.train INFO: Epoch[50] Iteration[100/186] Loss: 0.070, Acc: 0.995, Base Lr: 5.04e-03 +2022-11-04 22:05:24,960 transreid.train INFO: Epoch[50] Iteration[150/186] Loss: 0.075, Acc: 0.994, Base Lr: 5.04e-03 +2022-11-04 22:05:37,592 transreid.train INFO: Epoch 50 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:05:56,797 transreid.train INFO: Epoch[51] Iteration[50/186] Loss: 0.065, Acc: 0.993, Base Lr: 4.94e-03 +2022-11-04 22:06:15,969 transreid.train INFO: Epoch[51] Iteration[100/186] Loss: 0.071, Acc: 0.994, Base Lr: 4.94e-03 +2022-11-04 22:06:35,104 transreid.train INFO: Epoch[51] Iteration[150/186] Loss: 0.065, Acc: 0.994, Base Lr: 4.94e-03 +2022-11-04 22:06:47,754 transreid.train INFO: Epoch 51 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:07:06,934 transreid.train INFO: Epoch[52] Iteration[50/186] Loss: 0.074, Acc: 0.994, Base Lr: 4.84e-03 +2022-11-04 22:07:26,082 transreid.train INFO: Epoch[52] Iteration[100/186] Loss: 0.066, Acc: 0.994, Base Lr: 4.84e-03 +2022-11-04 22:07:45,263 transreid.train INFO: Epoch[52] Iteration[150/186] Loss: 0.058, Acc: 0.995, Base Lr: 4.84e-03 +2022-11-04 22:07:58,315 transreid.train INFO: Epoch 52 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:08:17,542 transreid.train INFO: Epoch[53] Iteration[50/186] Loss: 0.064, Acc: 0.995, Base Lr: 4.74e-03 +2022-11-04 22:08:36,740 transreid.train INFO: Epoch[53] Iteration[100/186] Loss: 0.060, Acc: 0.994, Base Lr: 4.74e-03 +2022-11-04 22:08:55,927 transreid.train INFO: Epoch[53] Iteration[150/186] Loss: 0.059, Acc: 0.995, Base Lr: 4.74e-03 +2022-11-04 22:09:08,247 transreid.train INFO: Epoch 53 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 22:09:27,448 transreid.train INFO: Epoch[54] Iteration[50/186] Loss: 0.062, Acc: 0.993, Base Lr: 4.63e-03 +2022-11-04 22:09:46,611 transreid.train INFO: Epoch[54] Iteration[100/186] Loss: 0.066, Acc: 0.993, Base Lr: 4.63e-03 +2022-11-04 22:10:05,802 transreid.train INFO: Epoch[54] Iteration[150/186] Loss: 0.065, Acc: 0.994, Base Lr: 4.63e-03 +2022-11-04 22:10:18,445 transreid.train INFO: Epoch 54 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 22:10:37,636 transreid.train INFO: Epoch[55] Iteration[50/186] Loss: 0.062, Acc: 0.997, Base Lr: 4.53e-03 +2022-11-04 22:10:56,805 transreid.train INFO: Epoch[55] Iteration[100/186] Loss: 0.059, Acc: 0.996, Base Lr: 4.53e-03 +2022-11-04 22:11:15,941 transreid.train INFO: Epoch[55] Iteration[150/186] Loss: 0.062, Acc: 0.995, Base Lr: 4.53e-03 +2022-11-04 22:11:28,597 transreid.train INFO: Epoch 55 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:11:47,769 transreid.train INFO: Epoch[56] Iteration[50/186] Loss: 0.069, Acc: 0.995, Base Lr: 4.43e-03 +2022-11-04 22:12:06,940 transreid.train INFO: Epoch[56] Iteration[100/186] Loss: 0.067, Acc: 0.995, Base Lr: 4.43e-03 +2022-11-04 22:12:26,094 transreid.train INFO: Epoch[56] Iteration[150/186] Loss: 0.067, Acc: 0.995, Base Lr: 4.43e-03 +2022-11-04 22:12:38,755 transreid.train INFO: Epoch 56 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:12:57,958 transreid.train INFO: Epoch[57] Iteration[50/186] Loss: 0.064, Acc: 0.995, Base Lr: 4.32e-03 +2022-11-04 22:13:17,086 transreid.train INFO: Epoch[57] Iteration[100/186] Loss: 0.072, Acc: 0.994, Base Lr: 4.32e-03 +2022-11-04 22:13:36,245 transreid.train INFO: Epoch[57] Iteration[150/186] Loss: 0.065, Acc: 0.994, Base Lr: 4.32e-03 +2022-11-04 22:13:48,884 transreid.train INFO: Epoch 57 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:14:08,022 transreid.train INFO: Epoch[58] Iteration[50/186] Loss: 0.061, Acc: 0.994, Base Lr: 4.22e-03 +2022-11-04 22:14:27,184 transreid.train INFO: Epoch[58] Iteration[100/186] Loss: 0.056, Acc: 0.996, Base Lr: 4.22e-03 +2022-11-04 22:14:46,320 transreid.train INFO: Epoch[58] Iteration[150/186] Loss: 0.053, Acc: 0.995, Base Lr: 4.22e-03 +2022-11-04 22:14:58,961 transreid.train INFO: Epoch 58 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:15:18,106 transreid.train INFO: Epoch[59] Iteration[50/186] Loss: 0.049, Acc: 0.997, Base Lr: 4.11e-03 +2022-11-04 22:15:37,234 transreid.train INFO: Epoch[59] Iteration[100/186] Loss: 0.057, Acc: 0.996, Base Lr: 4.11e-03 +2022-11-04 22:15:56,386 transreid.train INFO: Epoch[59] Iteration[150/186] Loss: 0.051, Acc: 0.996, Base Lr: 4.11e-03 +2022-11-04 22:16:09,008 transreid.train INFO: Epoch 59 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 22:16:28,182 transreid.train INFO: Epoch[60] Iteration[50/186] Loss: 0.049, Acc: 0.997, Base Lr: 4.01e-03 +2022-11-04 22:16:47,373 transreid.train INFO: Epoch[60] Iteration[100/186] Loss: 0.048, Acc: 0.995, Base Lr: 4.01e-03 +2022-11-04 22:17:06,521 transreid.train INFO: Epoch[60] Iteration[150/186] Loss: 0.049, Acc: 0.996, Base Lr: 4.01e-03 +2022-11-04 22:17:19,160 transreid.train INFO: Epoch 60 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:17:38,348 transreid.train INFO: Epoch[61] Iteration[50/186] Loss: 0.060, Acc: 0.994, Base Lr: 3.90e-03 +2022-11-04 22:17:57,494 transreid.train INFO: Epoch[61] Iteration[100/186] Loss: 0.057, Acc: 0.993, Base Lr: 3.90e-03 +2022-11-04 22:18:16,669 transreid.train INFO: Epoch[61] Iteration[150/186] Loss: 0.052, Acc: 0.995, Base Lr: 3.90e-03 +2022-11-04 22:18:29,710 transreid.train INFO: Epoch 61 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:18:48,874 transreid.train INFO: Epoch[62] Iteration[50/186] Loss: 0.050, Acc: 0.997, Base Lr: 3.80e-03 +2022-11-04 22:19:08,024 transreid.train INFO: Epoch[62] Iteration[100/186] Loss: 0.054, Acc: 0.996, Base Lr: 3.80e-03 +2022-11-04 22:19:27,200 transreid.train INFO: Epoch[62] Iteration[150/186] Loss: 0.055, Acc: 0.996, Base Lr: 3.80e-03 +2022-11-04 22:19:40,247 transreid.train INFO: Epoch 62 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:19:59,423 transreid.train INFO: Epoch[63] Iteration[50/186] Loss: 0.063, Acc: 0.990, Base Lr: 3.69e-03 +2022-11-04 22:20:18,584 transreid.train INFO: Epoch[63] Iteration[100/186] Loss: 0.054, Acc: 0.993, Base Lr: 3.69e-03 +2022-11-04 22:20:37,730 transreid.train INFO: Epoch[63] Iteration[150/186] Loss: 0.054, Acc: 0.994, Base Lr: 3.69e-03 +2022-11-04 22:20:50,767 transreid.train INFO: Epoch 63 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:21:09,908 transreid.train INFO: Epoch[64] Iteration[50/186] Loss: 0.065, Acc: 0.996, Base Lr: 3.59e-03 +2022-11-04 22:21:29,078 transreid.train INFO: Epoch[64] Iteration[100/186] Loss: 0.061, Acc: 0.995, Base Lr: 3.59e-03 +2022-11-04 22:21:48,233 transreid.train INFO: Epoch[64] Iteration[150/186] Loss: 0.057, Acc: 0.995, Base Lr: 3.59e-03 +2022-11-04 22:22:00,921 transreid.train INFO: Epoch 64 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:22:20,084 transreid.train INFO: Epoch[65] Iteration[50/186] Loss: 0.062, Acc: 0.995, Base Lr: 3.49e-03 +2022-11-04 22:22:39,235 transreid.train INFO: Epoch[65] Iteration[100/186] Loss: 0.053, Acc: 0.996, Base Lr: 3.49e-03 +2022-11-04 22:22:58,361 transreid.train INFO: Epoch[65] Iteration[150/186] Loss: 0.049, Acc: 0.996, Base Lr: 3.49e-03 +2022-11-04 22:23:11,367 transreid.train INFO: Epoch 65 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 22:23:30,504 transreid.train INFO: Epoch[66] Iteration[50/186] Loss: 0.048, Acc: 0.998, Base Lr: 3.38e-03 +2022-11-04 22:23:49,687 transreid.train INFO: Epoch[66] Iteration[100/186] Loss: 0.042, Acc: 0.997, Base Lr: 3.38e-03 +2022-11-04 22:24:08,842 transreid.train INFO: Epoch[66] Iteration[150/186] Loss: 0.044, Acc: 0.997, Base Lr: 3.38e-03 +2022-11-04 22:24:21,472 transreid.train INFO: Epoch 66 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:24:40,636 transreid.train INFO: Epoch[67] Iteration[50/186] Loss: 0.049, Acc: 0.995, Base Lr: 3.28e-03 +2022-11-04 22:24:59,766 transreid.train INFO: Epoch[67] Iteration[100/186] Loss: 0.043, Acc: 0.995, Base Lr: 3.28e-03 +2022-11-04 22:25:18,911 transreid.train INFO: Epoch[67] Iteration[150/186] Loss: 0.041, Acc: 0.996, Base Lr: 3.28e-03 +2022-11-04 22:25:31,906 transreid.train INFO: Epoch 67 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 22:25:51,079 transreid.train INFO: Epoch[68] Iteration[50/186] Loss: 0.053, Acc: 0.996, Base Lr: 3.18e-03 +2022-11-04 22:26:10,225 transreid.train INFO: Epoch[68] Iteration[100/186] Loss: 0.051, Acc: 0.995, Base Lr: 3.18e-03 +2022-11-04 22:26:29,370 transreid.train INFO: Epoch[68] Iteration[150/186] Loss: 0.045, Acc: 0.996, Base Lr: 3.18e-03 +2022-11-04 22:26:42,023 transreid.train INFO: Epoch 68 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:27:01,185 transreid.train INFO: Epoch[69] Iteration[50/186] Loss: 0.047, Acc: 0.995, Base Lr: 3.08e-03 +2022-11-04 22:27:20,300 transreid.train INFO: Epoch[69] Iteration[100/186] Loss: 0.052, Acc: 0.995, Base Lr: 3.08e-03 +2022-11-04 22:27:39,429 transreid.train INFO: Epoch[69] Iteration[150/186] Loss: 0.047, Acc: 0.996, Base Lr: 3.08e-03 +2022-11-04 22:27:52,063 transreid.train INFO: Epoch 69 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 22:28:11,233 transreid.train INFO: Epoch[70] Iteration[50/186] Loss: 0.045, Acc: 0.996, Base Lr: 2.97e-03 +2022-11-04 22:28:30,364 transreid.train INFO: Epoch[70] Iteration[100/186] Loss: 0.038, Acc: 0.997, Base Lr: 2.97e-03 +2022-11-04 22:28:49,501 transreid.train INFO: Epoch[70] Iteration[150/186] Loss: 0.034, Acc: 0.997, Base Lr: 2.97e-03 +2022-11-04 22:29:02,545 transreid.train INFO: Epoch 70 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:29:21,700 transreid.train INFO: Epoch[71] Iteration[50/186] Loss: 0.050, Acc: 0.996, Base Lr: 2.87e-03 +2022-11-04 22:29:40,877 transreid.train INFO: Epoch[71] Iteration[100/186] Loss: 0.040, Acc: 0.997, Base Lr: 2.87e-03 +2022-11-04 22:30:00,053 transreid.train INFO: Epoch[71] Iteration[150/186] Loss: 0.038, Acc: 0.997, Base Lr: 2.87e-03 +2022-11-04 22:30:13,076 transreid.train INFO: Epoch 71 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:30:32,239 transreid.train INFO: Epoch[72] Iteration[50/186] Loss: 0.040, Acc: 0.998, Base Lr: 2.77e-03 +2022-11-04 22:30:51,387 transreid.train INFO: Epoch[72] Iteration[100/186] Loss: 0.047, Acc: 0.997, Base Lr: 2.77e-03 +2022-11-04 22:31:10,549 transreid.train INFO: Epoch[72] Iteration[150/186] Loss: 0.042, Acc: 0.997, Base Lr: 2.77e-03 +2022-11-04 22:31:23,198 transreid.train INFO: Epoch 72 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:31:42,392 transreid.train INFO: Epoch[73] Iteration[50/186] Loss: 0.040, Acc: 0.997, Base Lr: 2.68e-03 +2022-11-04 22:32:01,519 transreid.train INFO: Epoch[73] Iteration[100/186] Loss: 0.036, Acc: 0.997, Base Lr: 2.68e-03 +2022-11-04 22:32:20,646 transreid.train INFO: Epoch[73] Iteration[150/186] Loss: 0.042, Acc: 0.996, Base Lr: 2.68e-03 +2022-11-04 22:32:33,275 transreid.train INFO: Epoch 73 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:32:52,454 transreid.train INFO: Epoch[74] Iteration[50/186] Loss: 0.040, Acc: 0.995, Base Lr: 2.58e-03 +2022-11-04 22:33:11,621 transreid.train INFO: Epoch[74] Iteration[100/186] Loss: 0.040, Acc: 0.997, Base Lr: 2.58e-03 +2022-11-04 22:33:30,768 transreid.train INFO: Epoch[74] Iteration[150/186] Loss: 0.036, Acc: 0.998, Base Lr: 2.58e-03 +2022-11-04 22:33:43,401 transreid.train INFO: Epoch 74 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:34:02,563 transreid.train INFO: Epoch[75] Iteration[50/186] Loss: 0.034, Acc: 0.997, Base Lr: 2.48e-03 +2022-11-04 22:34:21,718 transreid.train INFO: Epoch[75] Iteration[100/186] Loss: 0.031, Acc: 0.998, Base Lr: 2.48e-03 +2022-11-04 22:34:40,894 transreid.train INFO: Epoch[75] Iteration[150/186] Loss: 0.030, Acc: 0.998, Base Lr: 2.48e-03 +2022-11-04 22:34:53,532 transreid.train INFO: Epoch 75 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:35:12,662 transreid.train INFO: Epoch[76] Iteration[50/186] Loss: 0.035, Acc: 0.998, Base Lr: 2.38e-03 +2022-11-04 22:35:31,824 transreid.train INFO: Epoch[76] Iteration[100/186] Loss: 0.031, Acc: 0.998, Base Lr: 2.38e-03 +2022-11-04 22:35:51,000 transreid.train INFO: Epoch[76] Iteration[150/186] Loss: 0.029, Acc: 0.998, Base Lr: 2.38e-03 +2022-11-04 22:36:04,022 transreid.train INFO: Epoch 76 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:36:23,184 transreid.train INFO: Epoch[77] Iteration[50/186] Loss: 0.035, Acc: 0.998, Base Lr: 2.29e-03 +2022-11-04 22:36:42,347 transreid.train INFO: Epoch[77] Iteration[100/186] Loss: 0.042, Acc: 0.996, Base Lr: 2.29e-03 +2022-11-04 22:37:01,509 transreid.train INFO: Epoch[77] Iteration[150/186] Loss: 0.039, Acc: 0.997, Base Lr: 2.29e-03 +2022-11-04 22:37:14,542 transreid.train INFO: Epoch 77 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:37:33,731 transreid.train INFO: Epoch[78] Iteration[50/186] Loss: 0.038, Acc: 0.997, Base Lr: 2.20e-03 +2022-11-04 22:37:52,880 transreid.train INFO: Epoch[78] Iteration[100/186] Loss: 0.035, Acc: 0.998, Base Lr: 2.20e-03 +2022-11-04 22:38:12,052 transreid.train INFO: Epoch[78] Iteration[150/186] Loss: 0.035, Acc: 0.998, Base Lr: 2.20e-03 +2022-11-04 22:38:24,699 transreid.train INFO: Epoch 78 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:38:43,877 transreid.train INFO: Epoch[79] Iteration[50/186] Loss: 0.044, Acc: 0.997, Base Lr: 2.10e-03 +2022-11-04 22:39:03,014 transreid.train INFO: Epoch[79] Iteration[100/186] Loss: 0.040, Acc: 0.997, Base Lr: 2.10e-03 +2022-11-04 22:39:22,180 transreid.train INFO: Epoch[79] Iteration[150/186] Loss: 0.038, Acc: 0.997, Base Lr: 2.10e-03 +2022-11-04 22:39:35,203 transreid.train INFO: Epoch 79 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:39:54,344 transreid.train INFO: Epoch[80] Iteration[50/186] Loss: 0.032, Acc: 0.998, Base Lr: 2.01e-03 +2022-11-04 22:40:13,462 transreid.train INFO: Epoch[80] Iteration[100/186] Loss: 0.032, Acc: 0.997, Base Lr: 2.01e-03 +2022-11-04 22:40:32,628 transreid.train INFO: Epoch[80] Iteration[150/186] Loss: 0.032, Acc: 0.997, Base Lr: 2.01e-03 +2022-11-04 22:40:44,874 transreid.train INFO: Epoch 80 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 22:41:04,034 transreid.train INFO: Epoch[81] Iteration[50/186] Loss: 0.041, Acc: 0.997, Base Lr: 1.92e-03 +2022-11-04 22:41:23,199 transreid.train INFO: Epoch[81] Iteration[100/186] Loss: 0.038, Acc: 0.997, Base Lr: 1.92e-03 +2022-11-04 22:41:42,353 transreid.train INFO: Epoch[81] Iteration[150/186] Loss: 0.035, Acc: 0.997, Base Lr: 1.92e-03 +2022-11-04 22:41:55,377 transreid.train INFO: Epoch 81 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:42:14,536 transreid.train INFO: Epoch[82] Iteration[50/186] Loss: 0.050, Acc: 0.997, Base Lr: 1.83e-03 +2022-11-04 22:42:33,722 transreid.train INFO: Epoch[82] Iteration[100/186] Loss: 0.042, Acc: 0.997, Base Lr: 1.83e-03 +2022-11-04 22:42:52,907 transreid.train INFO: Epoch[82] Iteration[150/186] Loss: 0.039, Acc: 0.998, Base Lr: 1.83e-03 +2022-11-04 22:43:05,927 transreid.train INFO: Epoch 82 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:43:25,102 transreid.train INFO: Epoch[83] Iteration[50/186] Loss: 0.041, Acc: 0.997, Base Lr: 1.75e-03 +2022-11-04 22:43:44,246 transreid.train INFO: Epoch[83] Iteration[100/186] Loss: 0.034, Acc: 0.997, Base Lr: 1.75e-03 +2022-11-04 22:44:03,412 transreid.train INFO: Epoch[83] Iteration[150/186] Loss: 0.031, Acc: 0.997, Base Lr: 1.75e-03 +2022-11-04 22:44:15,666 transreid.train INFO: Epoch 83 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:44:34,877 transreid.train INFO: Epoch[84] Iteration[50/186] Loss: 0.036, Acc: 0.996, Base Lr: 1.66e-03 +2022-11-04 22:44:54,072 transreid.train INFO: Epoch[84] Iteration[100/186] Loss: 0.037, Acc: 0.996, Base Lr: 1.66e-03 +2022-11-04 22:45:13,249 transreid.train INFO: Epoch[84] Iteration[150/186] Loss: 0.034, Acc: 0.996, Base Lr: 1.66e-03 +2022-11-04 22:45:25,906 transreid.train INFO: Epoch 84 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 22:45:45,131 transreid.train INFO: Epoch[85] Iteration[50/186] Loss: 0.039, Acc: 0.997, Base Lr: 1.58e-03 +2022-11-04 22:46:04,358 transreid.train INFO: Epoch[85] Iteration[100/186] Loss: 0.038, Acc: 0.996, Base Lr: 1.58e-03 +2022-11-04 22:46:23,511 transreid.train INFO: Epoch[85] Iteration[150/186] Loss: 0.036, Acc: 0.997, Base Lr: 1.58e-03 +2022-11-04 22:46:36,196 transreid.train INFO: Epoch 85 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 22:46:55,370 transreid.train INFO: Epoch[86] Iteration[50/186] Loss: 0.051, Acc: 0.996, Base Lr: 1.50e-03 +2022-11-04 22:47:14,492 transreid.train INFO: Epoch[86] Iteration[100/186] Loss: 0.040, Acc: 0.997, Base Lr: 1.50e-03 +2022-11-04 22:47:33,645 transreid.train INFO: Epoch[86] Iteration[150/186] Loss: 0.038, Acc: 0.997, Base Lr: 1.50e-03 +2022-11-04 22:47:46,304 transreid.train INFO: Epoch 86 done. Time per batch: 0.383[s] Speed: 167.1[samples/s] +2022-11-04 22:48:05,460 transreid.train INFO: Epoch[87] Iteration[50/186] Loss: 0.038, Acc: 0.997, Base Lr: 1.42e-03 +2022-11-04 22:48:24,666 transreid.train INFO: Epoch[87] Iteration[100/186] Loss: 0.037, Acc: 0.997, Base Lr: 1.42e-03 +2022-11-04 22:48:43,863 transreid.train INFO: Epoch[87] Iteration[150/186] Loss: 0.035, Acc: 0.996, Base Lr: 1.42e-03 +2022-11-04 22:48:56,522 transreid.train INFO: Epoch 87 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 22:49:15,712 transreid.train INFO: Epoch[88] Iteration[50/186] Loss: 0.031, Acc: 0.999, Base Lr: 1.34e-03 +2022-11-04 22:49:34,855 transreid.train INFO: Epoch[88] Iteration[100/186] Loss: 0.029, Acc: 0.999, Base Lr: 1.34e-03 +2022-11-04 22:49:54,020 transreid.train INFO: Epoch[88] Iteration[150/186] Loss: 0.027, Acc: 0.998, Base Lr: 1.34e-03 +2022-11-04 22:50:06,669 transreid.train INFO: Epoch 88 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:50:25,845 transreid.train INFO: Epoch[89] Iteration[50/186] Loss: 0.026, Acc: 0.998, Base Lr: 1.26e-03 +2022-11-04 22:50:45,023 transreid.train INFO: Epoch[89] Iteration[100/186] Loss: 0.029, Acc: 0.998, Base Lr: 1.26e-03 +2022-11-04 22:51:04,194 transreid.train INFO: Epoch[89] Iteration[150/186] Loss: 0.028, Acc: 0.998, Base Lr: 1.26e-03 +2022-11-04 22:51:16,419 transreid.train INFO: Epoch 89 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:51:35,573 transreid.train INFO: Epoch[90] Iteration[50/186] Loss: 0.032, Acc: 0.998, Base Lr: 1.19e-03 +2022-11-04 22:51:54,725 transreid.train INFO: Epoch[90] Iteration[100/186] Loss: 0.028, Acc: 0.998, Base Lr: 1.19e-03 +2022-11-04 22:52:13,838 transreid.train INFO: Epoch[90] Iteration[150/186] Loss: 0.028, Acc: 0.998, Base Lr: 1.19e-03 +2022-11-04 22:52:26,480 transreid.train INFO: Epoch 90 done. Time per batch: 0.383[s] Speed: 167.2[samples/s] +2022-11-04 22:52:45,655 transreid.train INFO: Epoch[91] Iteration[50/186] Loss: 0.044, Acc: 0.997, Base Lr: 1.11e-03 +2022-11-04 22:53:04,829 transreid.train INFO: Epoch[91] Iteration[100/186] Loss: 0.033, Acc: 0.998, Base Lr: 1.11e-03 +2022-11-04 22:53:23,992 transreid.train INFO: Epoch[91] Iteration[150/186] Loss: 0.032, Acc: 0.997, Base Lr: 1.11e-03 +2022-11-04 22:53:37,020 transreid.train INFO: Epoch 91 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 22:53:56,176 transreid.train INFO: Epoch[92] Iteration[50/186] Loss: 0.033, Acc: 0.997, Base Lr: 1.04e-03 +2022-11-04 22:54:15,318 transreid.train INFO: Epoch[92] Iteration[100/186] Loss: 0.030, Acc: 0.998, Base Lr: 1.04e-03 +2022-11-04 22:54:34,507 transreid.train INFO: Epoch[92] Iteration[150/186] Loss: 0.028, Acc: 0.998, Base Lr: 1.04e-03 +2022-11-04 22:54:47,528 transreid.train INFO: Epoch 92 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 22:55:06,711 transreid.train INFO: Epoch[93] Iteration[50/186] Loss: 0.032, Acc: 0.998, Base Lr: 9.72e-04 +2022-11-04 22:55:25,876 transreid.train INFO: Epoch[93] Iteration[100/186] Loss: 0.031, Acc: 0.998, Base Lr: 9.72e-04 +2022-11-04 22:55:45,053 transreid.train INFO: Epoch[93] Iteration[150/186] Loss: 0.029, Acc: 0.998, Base Lr: 9.72e-04 +2022-11-04 22:55:58,122 transreid.train INFO: Epoch 93 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 22:56:17,333 transreid.train INFO: Epoch[94] Iteration[50/186] Loss: 0.035, Acc: 0.998, Base Lr: 9.06e-04 +2022-11-04 22:56:36,525 transreid.train INFO: Epoch[94] Iteration[100/186] Loss: 0.029, Acc: 0.998, Base Lr: 9.06e-04 +2022-11-04 22:56:55,761 transreid.train INFO: Epoch[94] Iteration[150/186] Loss: 0.027, Acc: 0.998, Base Lr: 9.06e-04 +2022-11-04 22:57:08,798 transreid.train INFO: Epoch 94 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 22:57:28,029 transreid.train INFO: Epoch[95] Iteration[50/186] Loss: 0.030, Acc: 0.998, Base Lr: 8.41e-04 +2022-11-04 22:57:47,255 transreid.train INFO: Epoch[95] Iteration[100/186] Loss: 0.026, Acc: 0.998, Base Lr: 8.41e-04 +2022-11-04 22:58:06,445 transreid.train INFO: Epoch[95] Iteration[150/186] Loss: 0.026, Acc: 0.998, Base Lr: 8.41e-04 +2022-11-04 22:58:19,111 transreid.train INFO: Epoch 95 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 22:58:38,316 transreid.train INFO: Epoch[96] Iteration[50/186] Loss: 0.028, Acc: 0.997, Base Lr: 7.78e-04 +2022-11-04 22:58:57,506 transreid.train INFO: Epoch[96] Iteration[100/186] Loss: 0.026, Acc: 0.998, Base Lr: 7.78e-04 +2022-11-04 22:59:16,747 transreid.train INFO: Epoch[96] Iteration[150/186] Loss: 0.026, Acc: 0.998, Base Lr: 7.78e-04 +2022-11-04 22:59:29,043 transreid.train INFO: Epoch 96 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 22:59:48,243 transreid.train INFO: Epoch[97] Iteration[50/186] Loss: 0.028, Acc: 0.998, Base Lr: 7.18e-04 +2022-11-04 23:00:07,424 transreid.train INFO: Epoch[97] Iteration[100/186] Loss: 0.029, Acc: 0.998, Base Lr: 7.18e-04 +2022-11-04 23:00:26,623 transreid.train INFO: Epoch[97] Iteration[150/186] Loss: 0.026, Acc: 0.998, Base Lr: 7.18e-04 +2022-11-04 23:00:39,666 transreid.train INFO: Epoch 97 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:00:58,840 transreid.train INFO: Epoch[98] Iteration[50/186] Loss: 0.025, Acc: 0.999, Base Lr: 6.60e-04 +2022-11-04 23:01:18,039 transreid.train INFO: Epoch[98] Iteration[100/186] Loss: 0.027, Acc: 0.999, Base Lr: 6.60e-04 +2022-11-04 23:01:37,225 transreid.train INFO: Epoch[98] Iteration[150/186] Loss: 0.028, Acc: 0.998, Base Lr: 6.60e-04 +2022-11-04 23:01:49,892 transreid.train INFO: Epoch 98 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 23:02:09,126 transreid.train INFO: Epoch[99] Iteration[50/186] Loss: 0.030, Acc: 0.998, Base Lr: 6.04e-04 +2022-11-04 23:02:28,295 transreid.train INFO: Epoch[99] Iteration[100/186] Loss: 0.032, Acc: 0.998, Base Lr: 6.04e-04 +2022-11-04 23:02:47,501 transreid.train INFO: Epoch[99] Iteration[150/186] Loss: 0.027, Acc: 0.998, Base Lr: 6.04e-04 +2022-11-04 23:03:00,562 transreid.train INFO: Epoch 99 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 23:03:19,753 transreid.train INFO: Epoch[100] Iteration[50/186] Loss: 0.028, Acc: 0.997, Base Lr: 5.51e-04 +2022-11-04 23:03:38,968 transreid.train INFO: Epoch[100] Iteration[100/186] Loss: 0.029, Acc: 0.998, Base Lr: 5.51e-04 +2022-11-04 23:03:58,156 transreid.train INFO: Epoch[100] Iteration[150/186] Loss: 0.027, Acc: 0.998, Base Lr: 5.51e-04 +2022-11-04 23:04:10,823 transreid.train INFO: Epoch 100 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:04:30,034 transreid.train INFO: Epoch[101] Iteration[50/186] Loss: 0.039, Acc: 0.997, Base Lr: 5.00e-04 +2022-11-04 23:04:49,179 transreid.train INFO: Epoch[101] Iteration[100/186] Loss: 0.031, Acc: 0.998, Base Lr: 5.00e-04 +2022-11-04 23:05:08,346 transreid.train INFO: Epoch[101] Iteration[150/186] Loss: 0.029, Acc: 0.998, Base Lr: 5.00e-04 +2022-11-04 23:05:20,979 transreid.train INFO: Epoch 101 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 23:05:40,194 transreid.train INFO: Epoch[102] Iteration[50/186] Loss: 0.028, Acc: 0.998, Base Lr: 4.51e-04 +2022-11-04 23:05:59,379 transreid.train INFO: Epoch[102] Iteration[100/186] Loss: 0.026, Acc: 0.998, Base Lr: 4.51e-04 +2022-11-04 23:06:18,612 transreid.train INFO: Epoch[102] Iteration[150/186] Loss: 0.027, Acc: 0.998, Base Lr: 4.51e-04 +2022-11-04 23:06:31,674 transreid.train INFO: Epoch 102 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 23:06:50,898 transreid.train INFO: Epoch[103] Iteration[50/186] Loss: 0.027, Acc: 0.999, Base Lr: 4.05e-04 +2022-11-04 23:07:10,104 transreid.train INFO: Epoch[103] Iteration[100/186] Loss: 0.027, Acc: 0.998, Base Lr: 4.05e-04 +2022-11-04 23:07:29,308 transreid.train INFO: Epoch[103] Iteration[150/186] Loss: 0.025, Acc: 0.998, Base Lr: 4.05e-04 +2022-11-04 23:07:41,983 transreid.train INFO: Epoch 103 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 23:08:01,208 transreid.train INFO: Epoch[104] Iteration[50/186] Loss: 0.032, Acc: 0.998, Base Lr: 3.61e-04 +2022-11-04 23:08:20,449 transreid.train INFO: Epoch[104] Iteration[100/186] Loss: 0.027, Acc: 0.998, Base Lr: 3.61e-04 +2022-11-04 23:08:39,681 transreid.train INFO: Epoch[104] Iteration[150/186] Loss: 0.026, Acc: 0.998, Base Lr: 3.61e-04 +2022-11-04 23:08:52,383 transreid.train INFO: Epoch 104 done. Time per batch: 0.385[s] Speed: 166.4[samples/s] +2022-11-04 23:09:11,627 transreid.train INFO: Epoch[105] Iteration[50/186] Loss: 0.028, Acc: 0.999, Base Lr: 3.20e-04 +2022-11-04 23:09:30,854 transreid.train INFO: Epoch[105] Iteration[100/186] Loss: 0.028, Acc: 0.999, Base Lr: 3.20e-04 +2022-11-04 23:09:50,105 transreid.train INFO: Epoch[105] Iteration[150/186] Loss: 0.026, Acc: 0.999, Base Lr: 3.20e-04 +2022-11-04 23:10:02,808 transreid.train INFO: Epoch 105 done. Time per batch: 0.385[s] Speed: 166.3[samples/s] +2022-11-04 23:10:22,043 transreid.train INFO: Epoch[106] Iteration[50/186] Loss: 0.020, Acc: 0.999, Base Lr: 2.81e-04 +2022-11-04 23:10:41,237 transreid.train INFO: Epoch[106] Iteration[100/186] Loss: 0.021, Acc: 0.999, Base Lr: 2.81e-04 +2022-11-04 23:11:00,439 transreid.train INFO: Epoch[106] Iteration[150/186] Loss: 0.020, Acc: 0.998, Base Lr: 2.81e-04 +2022-11-04 23:11:13,087 transreid.train INFO: Epoch 106 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:11:32,297 transreid.train INFO: Epoch[107] Iteration[50/186] Loss: 0.031, Acc: 0.998, Base Lr: 2.45e-04 +2022-11-04 23:11:51,483 transreid.train INFO: Epoch[107] Iteration[100/186] Loss: 0.030, Acc: 0.998, Base Lr: 2.45e-04 +2022-11-04 23:12:10,679 transreid.train INFO: Epoch[107] Iteration[150/186] Loss: 0.026, Acc: 0.998, Base Lr: 2.45e-04 +2022-11-04 23:12:23,332 transreid.train INFO: Epoch 107 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:12:42,528 transreid.train INFO: Epoch[108] Iteration[50/186] Loss: 0.034, Acc: 0.998, Base Lr: 2.11e-04 +2022-11-04 23:13:01,703 transreid.train INFO: Epoch[108] Iteration[100/186] Loss: 0.032, Acc: 0.999, Base Lr: 2.11e-04 +2022-11-04 23:13:20,904 transreid.train INFO: Epoch[108] Iteration[150/186] Loss: 0.028, Acc: 0.998, Base Lr: 2.11e-04 +2022-11-04 23:13:33,564 transreid.train INFO: Epoch 108 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 23:13:52,768 transreid.train INFO: Epoch[109] Iteration[50/186] Loss: 0.028, Acc: 0.998, Base Lr: 1.80e-04 +2022-11-04 23:14:11,968 transreid.train INFO: Epoch[109] Iteration[100/186] Loss: 0.026, Acc: 0.998, Base Lr: 1.80e-04 +2022-11-04 23:14:31,180 transreid.train INFO: Epoch[109] Iteration[150/186] Loss: 0.023, Acc: 0.998, Base Lr: 1.80e-04 +2022-11-04 23:14:43,839 transreid.train INFO: Epoch 109 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:15:03,026 transreid.train INFO: Epoch[110] Iteration[50/186] Loss: 0.029, Acc: 0.997, Base Lr: 1.52e-04 +2022-11-04 23:15:22,235 transreid.train INFO: Epoch[110] Iteration[100/186] Loss: 0.025, Acc: 0.998, Base Lr: 1.52e-04 +2022-11-04 23:15:41,443 transreid.train INFO: Epoch[110] Iteration[150/186] Loss: 0.022, Acc: 0.998, Base Lr: 1.52e-04 +2022-11-04 23:15:54,492 transreid.train INFO: Epoch 110 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:16:13,741 transreid.train INFO: Epoch[111] Iteration[50/186] Loss: 0.032, Acc: 0.997, Base Lr: 1.26e-04 +2022-11-04 23:16:32,925 transreid.train INFO: Epoch[111] Iteration[100/186] Loss: 0.031, Acc: 0.997, Base Lr: 1.26e-04 +2022-11-04 23:16:52,126 transreid.train INFO: Epoch[111] Iteration[150/186] Loss: 0.030, Acc: 0.997, Base Lr: 1.26e-04 +2022-11-04 23:17:04,810 transreid.train INFO: Epoch 111 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 23:17:24,025 transreid.train INFO: Epoch[112] Iteration[50/186] Loss: 0.027, Acc: 0.998, Base Lr: 1.03e-04 +2022-11-04 23:17:43,207 transreid.train INFO: Epoch[112] Iteration[100/186] Loss: 0.026, Acc: 0.998, Base Lr: 1.03e-04 +2022-11-04 23:18:02,395 transreid.train INFO: Epoch[112] Iteration[150/186] Loss: 0.022, Acc: 0.998, Base Lr: 1.03e-04 +2022-11-04 23:18:15,063 transreid.train INFO: Epoch 112 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:18:34,280 transreid.train INFO: Epoch[113] Iteration[50/186] Loss: 0.029, Acc: 0.998, Base Lr: 8.28e-05 +2022-11-04 23:18:53,456 transreid.train INFO: Epoch[113] Iteration[100/186] Loss: 0.028, Acc: 0.998, Base Lr: 8.28e-05 +2022-11-04 23:19:12,653 transreid.train INFO: Epoch[113] Iteration[150/186] Loss: 0.027, Acc: 0.998, Base Lr: 8.28e-05 +2022-11-04 23:19:25,308 transreid.train INFO: Epoch 113 done. Time per batch: 0.384[s] Speed: 166.7[samples/s] +2022-11-04 23:19:44,474 transreid.train INFO: Epoch[114] Iteration[50/186] Loss: 0.025, Acc: 0.998, Base Lr: 6.51e-05 +2022-11-04 23:20:03,676 transreid.train INFO: Epoch[114] Iteration[100/186] Loss: 0.026, Acc: 0.998, Base Lr: 6.51e-05 +2022-11-04 23:20:22,841 transreid.train INFO: Epoch[114] Iteration[150/186] Loss: 0.024, Acc: 0.998, Base Lr: 6.51e-05 +2022-11-04 23:20:35,511 transreid.train INFO: Epoch 114 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 23:20:54,743 transreid.train INFO: Epoch[115] Iteration[50/186] Loss: 0.027, Acc: 0.999, Base Lr: 5.02e-05 +2022-11-04 23:21:13,916 transreid.train INFO: Epoch[115] Iteration[100/186] Loss: 0.028, Acc: 0.998, Base Lr: 5.02e-05 +2022-11-04 23:21:33,101 transreid.train INFO: Epoch[115] Iteration[150/186] Loss: 0.027, Acc: 0.999, Base Lr: 5.02e-05 +2022-11-04 23:21:46,124 transreid.train INFO: Epoch 115 done. Time per batch: 0.384[s] Speed: 166.8[samples/s] +2022-11-04 23:22:05,330 transreid.train INFO: Epoch[116] Iteration[50/186] Loss: 0.032, Acc: 0.999, Base Lr: 3.79e-05 +2022-11-04 23:22:24,481 transreid.train INFO: Epoch[116] Iteration[100/186] Loss: 0.026, Acc: 0.999, Base Lr: 3.79e-05 +2022-11-04 23:22:43,657 transreid.train INFO: Epoch[116] Iteration[150/186] Loss: 0.025, Acc: 0.998, Base Lr: 3.79e-05 +2022-11-04 23:22:56,664 transreid.train INFO: Epoch 116 done. Time per batch: 0.383[s] Speed: 166.9[samples/s] +2022-11-04 23:23:15,847 transreid.train INFO: Epoch[117] Iteration[50/186] Loss: 0.026, Acc: 0.997, Base Lr: 2.83e-05 +2022-11-04 23:23:34,985 transreid.train INFO: Epoch[117] Iteration[100/186] Loss: 0.025, Acc: 0.997, Base Lr: 2.83e-05 +2022-11-04 23:23:54,157 transreid.train INFO: Epoch[117] Iteration[150/186] Loss: 0.024, Acc: 0.998, Base Lr: 2.83e-05 +2022-11-04 23:24:06,806 transreid.train INFO: Epoch 117 done. Time per batch: 0.383[s] Speed: 167.0[samples/s] +2022-11-04 23:24:26,013 transreid.train INFO: Epoch[118] Iteration[50/186] Loss: 0.024, Acc: 0.998, Base Lr: 2.15e-05 +2022-11-04 23:24:45,207 transreid.train INFO: Epoch[118] Iteration[100/186] Loss: 0.021, Acc: 0.999, Base Lr: 2.15e-05 +2022-11-04 23:25:04,399 transreid.train INFO: Epoch[118] Iteration[150/186] Loss: 0.021, Acc: 0.998, Base Lr: 2.15e-05 +2022-11-04 23:25:17,503 transreid.train INFO: Epoch 118 done. Time per batch: 0.384[s] Speed: 166.6[samples/s] +2022-11-04 23:25:37,092 transreid.train INFO: Epoch[119] Iteration[50/186] Loss: 0.023, Acc: 0.999, Base Lr: 1.74e-05 +2022-11-04 23:25:56,274 transreid.train INFO: Epoch[119] Iteration[100/186] Loss: 0.025, Acc: 0.998, Base Lr: 1.74e-05 +2022-11-04 23:26:15,416 transreid.train INFO: Epoch[119] Iteration[150/186] Loss: 0.024, Acc: 0.998, Base Lr: 1.74e-05 +2022-11-04 23:26:28,089 transreid.train INFO: Epoch 119 done. Time per batch: 0.386[s] Speed: 165.9[samples/s] +2022-11-04 23:26:47,292 transreid.train INFO: Epoch[120] Iteration[50/186] Loss: 0.025, Acc: 0.998, Base Lr: 1.60e-05 +2022-11-04 23:27:06,461 transreid.train INFO: Epoch[120] Iteration[100/186] Loss: 0.025, Acc: 0.997, Base Lr: 1.60e-05 +2022-11-04 23:27:25,627 transreid.train INFO: Epoch[120] Iteration[150/186] Loss: 0.024, Acc: 0.997, Base Lr: 1.60e-05 +2022-11-04 23:27:38,276 transreid.train INFO: Epoch 120 done. Time per batch: 0.384[s] Speed: 166.9[samples/s] +2022-11-04 23:28:56,140 transreid.train INFO: Validation Results - Epoch: 120 +2022-11-04 23:28:56,140 transreid.train INFO: mAP: 81.9% +2022-11-04 23:28:56,140 transreid.train INFO: CMC curve, Rank-1 :92.8% +2022-11-04 23:28:56,140 transreid.train INFO: CMC curve, Rank-5 :97.5% +2022-11-04 23:28:56,140 transreid.train INFO: CMC curve, Rank-10 :98.4% diff --git a/configs/Market/swin.yml b/configs/Market/swin.yml new file mode 100644 index 000000000..fa9024455 --- /dev/null +++ b/configs/Market/swin.yml @@ -0,0 +1,55 @@ +MODEL: + PRETRAIN_CHOICE: 'imagenet' + PRETRAIN_PATH: 'E:\\reid\\swin_base_patch4_window7_224_22k.pth' + METRIC_LOSS_TYPE: 'triplet' + IF_LABELSMOOTH: 'off' + IF_WITH_CENTER: 'no' + NAME: 'transformer' + NO_MARGIN: True + DEVICE_ID: ('0') + TRANSFORMER_TYPE: 'swin' + STRIDE_SIZE: [16, 16] + +INPUT: + SIZE_TRAIN: [256, 128] + SIZE_TEST: [256, 128] + PROB: 0.5 # random horizontal flip + RE_PROB: 0.5 # random erasing + PADDING: 10 + PIXEL_MEAN: [0.5, 0.5, 0.5] + PIXEL_STD: [0.5, 0.5, 0.5] + +DATASETS: + NAMES: ('market1501') + ROOT_DIR: ('E:\\TransReID-main') + +DATALOADER: + SAMPLER: 'softmax_triplet' + NUM_INSTANCE: 4 + NUM_WORKERS: 0 + +SOLVER: + OPTIMIZER_NAME: 'SGD' + MAX_EPOCHS: 120 + BASE_LR: 0.008 + IMS_PER_BATCH: 16 + WARMUP_METHOD: 'linear' + LARGE_FC_LR: False + CHECKPOINT_PERIOD: 120 + LOG_PERIOD: 50 + EVAL_PERIOD: 120 + WEIGHT_DECAY: 1e-4 + WEIGHT_DECAY_BIAS: 1e-4 + BIAS_LR_FACTOR: 2 + +TEST: + EVAL: True + IMS_PER_BATCH: 256 + RE_RANKING: False + WEIGHT: 'D:\\TransReID\\0321_market_vit_base\\transformer_120.pth' + NECK_FEAT: 'before' + FEAT_NORM: 'yes' + +OUTPUT_DIR: 'D:\\TransReID\\0321_market_vit_base' + + diff --git a/configs/Market/vit_base.yml b/configs/Market/vit_base.yml index 3af53254b..f344eca72 100644 --- a/configs/Market/vit_base.yml +++ b/configs/Market/vit_base.yml @@ -1,18 +1,18 @@ MODEL: PRETRAIN_CHOICE: 'imagenet' - PRETRAIN_PATH: '/home/heshuting/.cache/torch/checkpoints/jx_vit_base_p16_224-80ecf9dd.pth' + PRETRAIN_PATH: 'E:\\python_prj\\jx_vit_base_p16_224-80ecf9dd.pth' METRIC_LOSS_TYPE: 'triplet' IF_LABELSMOOTH: 'off' IF_WITH_CENTER: 'no' NAME: 'transformer' NO_MARGIN: True - DEVICE_ID: ('7') + DEVICE_ID: ('0') TRANSFORMER_TYPE: 'vit_base_patch16_224_TransReID' STRIDE_SIZE: [16, 16] INPUT: - SIZE_TRAIN: [256, 128] - SIZE_TEST: [256, 128] + SIZE_TRAIN: [256, 256] + SIZE_TEST: [256, 256] PROB: 0.5 # random horizontal flip RE_PROB: 0.5 # random erasing PADDING: 10 @@ -21,18 +21,18 @@ INPUT: DATASETS: NAMES: ('market1501') - ROOT_DIR: ('../../data') + ROOT_DIR: ('E:\\python_prj') DATALOADER: SAMPLER: 'softmax_triplet' - NUM_INSTANCE: 4 - NUM_WORKERS: 8 + NUM_INSTANCE: 1 + NUM_WORKERS: 0 SOLVER: OPTIMIZER_NAME: 'SGD' MAX_EPOCHS: 120 BASE_LR: 0.008 - IMS_PER_BATCH: 64 + IMS_PER_BATCH: 2 WARMUP_METHOD: 'linear' LARGE_FC_LR: False CHECKPOINT_PERIOD: 120 @@ -46,10 +46,10 @@ TEST: EVAL: True IMS_PER_BATCH: 256 RE_RANKING: False - WEIGHT: '../logs/0321_market_vit_base/transformer_120.pth' + WEIGHT: 'E:\\python_prj\\log\\transformer_120.pth' NECK_FEAT: 'before' FEAT_NORM: 'yes' -OUTPUT_DIR: '../logs/0321_market_vit_base' +OUTPUT_DIR: 'E:\\python_prj\\log' diff --git a/model/backbones/model.py b/model/backbones/model.py new file mode 100644 index 000000000..e93f4a4c7 --- /dev/null +++ b/model/backbones/model.py @@ -0,0 +1,720 @@ +""" Swin Transformer +A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` + - https://arxiv.org/pdf/2103.14030 + +Code/weights from https://github.com/microsoft/Swin-Transformer + +""" + +import torch +import torch.nn as nn +import torch.nn.functional as F +import torch.utils.checkpoint as checkpoint +import numpy as np +from typing import Optional +from einops.layers.torch import Rearrange, Reduce + +def drop_path_f(x, drop_prob: float = 0., training: bool = False): + """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). + + This is the same as the DropConnect impl I created for EfficientNet, etc networks, however, + the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper... + See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for + changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use + 'survival rate' as the argument. + + """ + if drop_prob == 0. or not training: + return x + keep_prob = 1 - drop_prob + shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets + random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device) + random_tensor.floor_() # binarize + output = x.div(keep_prob) * random_tensor + return output + + +class DropPath(nn.Module): + """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). + """ + def __init__(self, drop_prob=None): + super(DropPath, self).__init__() + self.drop_prob = drop_prob + + def forward(self, x): + return drop_path_f(x, self.drop_prob, self.training) + + +def window_partition(x, window_size: int): + """ + 将feature map按照window_size划分成一个个没有重叠的window + Args: + x: (B, H, W, C) + window_size (int): window size(M) + + Returns: + windows: (num_windows*B, window_size, window_size, C) + """ + B, H, W, C = x.shape + x = x.view(B, H // window_size, window_size, W // window_size, window_size, C) + # permute: [B, H//Mh, Mh, W//Mw, Mw, C] -> [B, H//Mh, W//Mh, Mw, Mw, C] + # view: [B, H//Mh, W//Mw, Mh, Mw, C] -> [B*num_windows, Mh, Mw, C] + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows + + +def window_reverse(windows, window_size: int, H: int, W: int): + """ + 将一个个window还原成一个feature map + Args: + windows: (num_windows*B, window_size, window_size, C) + window_size (int): Window size(M) + H (int): Height of image + W (int): Width of image + + Returns: + x: (B, H, W, C) + """ + B = int(windows.shape[0] / (H * W / window_size / window_size)) + # view: [B*num_windows, Mh, Mw, C] -> [B, H//Mh, W//Mw, Mh, Mw, C] + x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1) + # permute: [B, H//Mh, W//Mw, Mh, Mw, C] -> [B, H//Mh, Mh, W//Mw, Mw, C] + # view: [B, H//Mh, Mh, W//Mw, Mw, C] -> [B, H, W, C] + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1) + return x + + +class PatchEmbed(nn.Module): + """ + 2D Image to Patch Embedding + """ + def __init__(self, patch_size=4, in_c=3, embed_dim=96, norm_layer=None): + super().__init__() + patch_size = (patch_size, patch_size) + self.patch_size = patch_size + self.in_chans = in_c + self.embed_dim = embed_dim + self.proj = nn.Conv2d(in_c, embed_dim, kernel_size=patch_size, stride=patch_size) + self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity() + + def forward(self, x): + _, _, H, W = x.shape + + # padding + # 如果输入图片的H，W不是patch_size的整数倍，需要进行padding + pad_input = (H % self.patch_size[0] != 0) or (W % self.patch_size[1] != 0) + if pad_input: + # to pad the last 3 dimensions, + # (W_left, W_right, H_top,H_bottom, C_front, C_back) + x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1], + 0, self.patch_size[0] - H % self.patch_size[0], + 0, 0)) + + # 下采样patch_size倍 + x = self.proj(x) + _, _, H, W = x.shape + # flatten: [B, C, H, W] -> [B, C, HW] + # transpose: [B, C, HW] -> [B, HW, C] + x = x.flatten(2).transpose(1, 2) + x = self.norm(x) + return x, H, W + + +class PatchMerging(nn.Module): + r""" Patch Merging Layer. + + Args: + dim (int): Number of input channels. + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, dim, norm_layer=nn.LayerNorm): + super().__init__() + self.dim = dim + self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) + self.norm = norm_layer(4 * dim) + + def forward(self, x, H, W): + """ + x: B, H*W, C + """ + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + + x = x.view(B, H, W, C) + + # padding + # 如果输入feature map的H，W不是2的整数倍，需要进行padding + pad_input = (H % 2 == 1) or (W % 2 == 1) + if pad_input: + # to pad the last 3 dimensions, starting from the last dimension and moving forward. + # (C_front, C_back, W_left, W_right, H_top, H_bottom) + # 注意这里的Tensor通道是[B, H, W, C]，所以会和官方文档有些不同 + x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2)) + + x0 = x[:, 0::2, 0::2, :] # [B, H/2, W/2, C] + x1 = x[:, 1::2, 0::2, :] # [B, H/2, W/2, C] + x2 = x[:, 0::2, 1::2, :] # [B, H/2, W/2, C] + x3 = x[:, 1::2, 1::2, :] # [B, H/2, W/2, C] + x = torch.cat([x0, x1, x2, x3], -1) # [B, H/2, W/2, 4*C] + x = x.view(B, -1, 4 * C) # [B, H/2*W/2, 4*C] + + x = self.norm(x) + x = self.reduction(x) # [B, H/2*W/2, 2*C] + + return x + + +class Mlp(nn.Module): + """ MLP as used in Vision Transformer, MLP-Mixer and related networks + """ + def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + + self.fc1 = nn.Linear(in_features, hidden_features) + self.act = act_layer() + self.drop1 = nn.Dropout(drop) + self.fc2 = nn.Linear(hidden_features, out_features) + self.drop2 = nn.Dropout(drop) + + def forward(self, x): + x = self.fc1(x) + x = self.act(x) + x = self.drop1(x) + x = self.fc2(x) + x = self.drop2(x) + return x + + +class WindowAttention(nn.Module): + r""" Window based multi-head self attention (W-MSA) module with relative position bias. + It supports both of shifted and non-shifted window. + + Args: + dim (int): Number of input channels. + window_size (tuple[int]): The height and width of the window. + num_heads (int): Number of attention heads. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0 + proj_drop (float, optional): Dropout ratio of output. Default: 0.0 + """ + + def __init__(self, dim, window_size, num_heads, qkv_bias=True, attn_drop=0., proj_drop=0.): + + super().__init__() + self.dim = dim + self.window_size = window_size # [Mh, Mw] + self.num_heads = num_heads + head_dim = dim // num_heads + self.scale = head_dim ** -0.5 + + # define a parameter table of relative position bias + self.relative_position_bias_table = nn.Parameter( + torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)) # [2*Mh-1 * 2*Mw-1, nH] + + # get pair-wise relative position index for each token inside the window + coords_h = torch.arange(self.window_size[0]) + coords_w = torch.arange(self.window_size[1]) + coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # [2, Mh, Mw] + coords_flatten = torch.flatten(coords, 1) # [2, Mh*Mw] + # [2, Mh*Mw, 1] - [2, 1, Mh*Mw] + relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # [2, Mh*Mw, Mh*Mw] + relative_coords = relative_coords.permute(1, 2, 0).contiguous() # [Mh*Mw, Mh*Mw, 2] + relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0 + relative_coords[:, :, 1] += self.window_size[1] - 1 + relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 + relative_position_index = relative_coords.sum(-1) # [Mh*Mw, Mh*Mw] + self.register_buffer("relative_position_index", relative_position_index) + + self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) + self.attn_drop = nn.Dropout(attn_drop) + self.proj = nn.Linear(dim, dim) + self.proj_drop = nn.Dropout(proj_drop) + + nn.init.trunc_normal_(self.relative_position_bias_table, std=.02) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask: Optional[torch.Tensor] = None): + """ + Args: + x: input features with shape of (num_windows*B, Mh*Mw, C) + mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None + """ + # [batch_size*num_windows, Mh*Mw, total_embed_dim] + B_, N, C = x.shape + # qkv(): -> [batch_size*num_windows, Mh*Mw, 3 * total_embed_dim] + # reshape: -> [batch_size*num_windows, Mh*Mw, 3, num_heads, embed_dim_per_head] + # permute: -> [3, batch_size*num_windows, num_heads, Mh*Mw, embed_dim_per_head] + qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + # [batch_size*num_windows, num_heads, Mh*Mw, embed_dim_per_head] + q, k, v = qkv.unbind(0) # make torchscript happy (cannot use tensor as tuple) + + # transpose: -> [batch_size*num_windows, num_heads, embed_dim_per_head, Mh*Mw] + # @: multiply -> [batch_size*num_windows, num_heads, Mh*Mw, Mh*Mw] + q = q * self.scale + attn = (q @ k.transpose(-2, -1)) + + # relative_position_bias_table.view: [Mh*Mw*Mh*Mw,nH] -> [Mh*Mw,Mh*Mw,nH] + relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view( + self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) + relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # [nH, Mh*Mw, Mh*Mw] + attn = attn + relative_position_bias.unsqueeze(0) + + if mask is not None: + # mask: [nW, Mh*Mw, Mh*Mw] + nW = mask.shape[0] # num_windows + # attn.view: [batch_size, num_windows, num_heads, Mh*Mw, Mh*Mw] + # mask.unsqueeze: [1, nW, 1, Mh*Mw, Mh*Mw] + attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) + attn = attn.view(-1, self.num_heads, N, N) + attn = self.softmax(attn) + else: + attn = self.softmax(attn) + + attn = self.attn_drop(attn) + + # @: multiply -> [batch_size*num_windows, num_heads, Mh*Mw, embed_dim_per_head] + # transpose: -> [batch_size*num_windows, Mh*Mw, num_heads, embed_dim_per_head] + # reshape: -> [batch_size*num_windows, Mh*Mw, total_embed_dim] + x = (attn @ v).transpose(1, 2).reshape(B_, N, C) + x = self.proj(x) + x = self.proj_drop(x) + return x + + +class SwinTransformerBlock(nn.Module): + r""" Swin Transformer Block. + + Args: + dim (int): Number of input channels. + num_heads (int): Number of attention heads. + window_size (int): Window size. + shift_size (int): Shift size for SW-MSA. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float, optional): Stochastic depth rate. Default: 0.0 + act_layer (nn.Module, optional): Activation layer. Default: nn.GELU + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, dim, num_heads, window_size=7, shift_size=0, + mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., drop_path=0., + act_layer=nn.GELU, norm_layer=nn.LayerNorm): + super().__init__() + self.dim = dim + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + self.mlp_ratio = mlp_ratio + assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size" + + self.norm1 = norm_layer(dim) + self.attn = WindowAttention( + dim, window_size=(self.window_size, self.window_size), num_heads=num_heads, qkv_bias=qkv_bias, + attn_drop=attn_drop, proj_drop=drop) + + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + self.norm2 = norm_layer(dim) + mlp_hidden_dim = int(dim * mlp_ratio) + self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop) + + def forward(self, x, attn_mask): + H, W = self.H, self.W + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + + shortcut = x + x = self.norm1(x) + x = x.view(B, H, W, C) + + # pad feature maps to multiples of window size + # 把feature map给pad到window size的整数倍 + pad_l = pad_t = 0 + pad_r = (self.window_size - W % self.window_size) % self.window_size + pad_b = (self.window_size - H % self.window_size) % self.window_size + x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b)) + _, Hp, Wp, _ = x.shape + + # cyclic shift + if self.shift_size > 0: + shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) + else: + shifted_x = x + attn_mask = None + + # partition windows + x_windows = window_partition(shifted_x, self.window_size) # [nW*B, Mh, Mw, C] + x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # [nW*B, Mh*Mw, C] + + # W-MSA/SW-MSA + attn_windows = self.attn(x_windows, mask=attn_mask) # [nW*B, Mh*Mw, C] + + # merge windows + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) # [nW*B, Mh, Mw, C] + shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp) # [B, H', W', C] + + # reverse cyclic shift + if self.shift_size > 0: + x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) + else: + x = shifted_x + + if pad_r > 0 or pad_b > 0: + # 把前面pad的数据移除掉 + x = x[:, :H, :W, :].contiguous() + + x = x.view(B, H * W, C) + + # FFN + x = shortcut + self.drop_path(x) + x = x + self.drop_path(self.mlp(self.norm2(x))) + + return x + + +class BasicLayer(nn.Module): + """ + A basic Swin Transformer layer for one stage. + + Args: + dim (int): Number of input channels. + depth (int): Number of blocks. + num_heads (int): Number of attention heads. + window_size (int): Local window size. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0 + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + """ + + def __init__(self, dim, depth, num_heads, window_size, + mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., + drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False): + super().__init__() + self.dim = dim + self.depth = depth + self.window_size = window_size + self.use_checkpoint = use_checkpoint + self.shift_size = window_size // 2 + + # build blocks + self.blocks = nn.ModuleList([ + SwinTransformerBlock( + dim=dim, + num_heads=num_heads, + window_size=window_size, + shift_size=0 if (i % 2 == 0) else self.shift_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer) + for i in range(depth)]) + + self.conv1_1 = nn.Conv2d(dim * 2, dim, 1, 1, 0, bias=True) + + self.conv_block = nn.Sequential( + nn.Conv2d(dim, dim, 3, 1, 1, bias=False), + nn.ReLU(True), + nn.Conv2d(dim, dim, 3, 1, 1, bias=False) + ) + + # patch merging layer + if downsample is not None: + self.downsample = downsample(dim=dim, norm_layer=norm_layer) + else: + self.downsample = None + + def create_mask(self, x, H, W): + # calculate attention mask for SW-MSA + # 保证Hp和Wp是window_size的整数倍 + Hp = int(np.ceil(H / self.window_size)) * self.window_size + Wp = int(np.ceil(W / self.window_size)) * self.window_size + # 拥有和feature map一样的通道排列顺序，方便后续window_partition + img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device) # [1, Hp, Wp, 1] + h_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + w_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + cnt = 0 + for h in h_slices: + for w in w_slices: + img_mask[:, h, w, :] = cnt + cnt += 1 + + mask_windows = window_partition(img_mask, self.window_size) # [nW, Mh, Mw, 1] + mask_windows = mask_windows.view(-1, self.window_size * self.window_size) # [nW, Mh*Mw] + attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) # [nW, 1, Mh*Mw] - [nW, Mh*Mw, 1] + # [nW, Mh*Mw, Mh*Mw] + attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) + return attn_mask + + def forward(self, x, H, W): + attn_mask = self.create_mask(x, H, W) # [nW, Mh*Mw, Mh*Mw] + trans_x = x + conv_x = x + B, L, C = x.shape + conv_x = conv_x.view(B, H, W, C) + conv_x = Rearrange('b h w c -> b c h w')(conv_x) + + x = x.view(B, H, W, C) + x = Rearrange('b h w c -> b c h w')(x) + for blk in self.blocks: + blk.H, blk.W = H, W + if not torch.jit.is_scripting() and self.use_checkpoint: + x = checkpoint.checkpoint(blk, x, attn_mask) + else: + trans_x = blk(trans_x, attn_mask) + trans_x = trans_x.view(B, H, W, C) + trans_x = Rearrange('b h w c -> b c h w')(trans_x) + conv_x = self.conv_block(conv_x) + conv_x + res = self.conv1_1(torch.cat((conv_x, trans_x), dim=1)) + x = x + res + trans_x = Rearrange('b c h w -> b h w c')(trans_x) + trans_x = trans_x.view(B, H*W, C) + + x = Rearrange('b c h w -> b h w c')(x) + x = x.view(B, H*W, C) + if self.downsample is not None: + x = self.downsample(x, H, W) + H, W = (H + 1) // 2, (W + 1) // 2 + + return x, H, W + + +class SwinTransformer(nn.Module): + r""" Swin Transformer + A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` - + https://arxiv.org/pdf/2103.14030 + + Args: + patch_size (int | tuple(int)): Patch size. Default: 4 + in_chans (int): Number of input image channels. Default: 3 + num_classes (int): Number of classes for classification head. Default: 1000 + embed_dim (int): Patch embedding dimension. Default: 96 + depths (tuple(int)): Depth of each Swin Transformer layer. + num_heads (tuple(int)): Number of attention heads in different layers. + window_size (int): Window size. Default: 7 + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4 + qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True + drop_rate (float): Dropout rate. Default: 0 + attn_drop_rate (float): Attention dropout rate. Default: 0 + drop_path_rate (float): Stochastic depth rate. Default: 0.1 + norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. + patch_norm (bool): If True, add normalization after patch embedding. Default: True + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False + """ + + def __init__(self, patch_size=4, in_chans=3, num_classes=1000, + embed_dim=96, depths=(2, 2, 6, 2), num_heads=(3, 6, 12, 24), + window_size=7, mlp_ratio=4., qkv_bias=True, + drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1, + norm_layer=nn.LayerNorm, patch_norm=True, + use_checkpoint=False, **kwargs): + super().__init__() + + self.num_classes = num_classes + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.patch_norm = patch_norm + # stage4输出特征矩阵的channels + self.num_features = int(embed_dim * 2 ** (self.num_layers - 1)) + self.mlp_ratio = mlp_ratio + + # split image into non-overlapping patches + self.patch_embed = PatchEmbed( + patch_size=patch_size, in_c=in_chans, embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None) + self.pos_drop = nn.Dropout(p=drop_rate) + + # stochastic depth + dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule + + # build layers + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + # 注意这里构建的stage和论文图中有些差异 + # 这里的stage不包含该stage的patch_merging层，包含的是下个stage的 + layers = BasicLayer(dim=int(embed_dim * 2 ** i_layer), + depth=depths[i_layer], + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=self.mlp_ratio, + qkv_bias=qkv_bias, + drop=drop_rate, + attn_drop=attn_drop_rate, + drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], + norm_layer=norm_layer, + downsample=PatchMerging if (i_layer < self.num_layers - 1) else None, + use_checkpoint=use_checkpoint) + self.layers.append(layers) + + self.norm = norm_layer(self.num_features) + self.avgpool = nn.AdaptiveAvgPool1d(1) + self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity() + + self.apply(self._init_weights) + + def _init_weights(self, m): + if isinstance(m, nn.Linear): + nn.init.trunc_normal_(m.weight, std=.02) + if isinstance(m, nn.Linear) and m.bias is not None: + nn.init.constant_(m.bias, 0) + elif isinstance(m, nn.LayerNorm): + nn.init.constant_(m.bias, 0) + nn.init.constant_(m.weight, 1.0) + + def forward(self, x, label=None, cam_label= None, view_label=None): + # x: [B, L, C] + x, H, W = self.patch_embed(x) + x = self.pos_drop(x) + + for layer in self.layers: + x, H, W = layer(x, H, W) + + x = self.norm(x) # [B, L, C] + x = self.avgpool(x.transpose(1, 2)) # [B, C, 1] + x = torch.flatten(x, 1) + x = self.head(x) + + # #修改 + # feat = self.bottleneck(x) + # if self.training: + # if self.ID_LOSS_TYPE in ('arcface', 'cosface', 'amsoftmax', 'circle'): + # cls_score = self.classifier(feat, label) + # else: + # cls_score = self.classifier(feat) + # + # return cls_score, x # global feature for triplet loss + # else: + # if self.neck_feat == 'after': + # # print("Test with feature after BN") + # return feat + # else: + # # print("Test with feature before BN") + # return x + + return x + + +def swin_tiny_patch4_window7_224(num_classes: int = 768, **kwargs): + # trained ImageNet-1K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_tiny_patch4_window7_224.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=7, + embed_dim=96, + depths=(2, 2, 6, 2), + num_heads=(3, 6, 12, 24), + num_classes=num_classes, + **kwargs) + return model + + +def swin_small_patch4_window7_224(num_classes: int = 1000, **kwargs): + # trained ImageNet-1K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_small_patch4_window7_224.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=7, + embed_dim=96, + depths=(2, 2, 18, 2), + num_heads=(3, 6, 12, 24), + num_classes=num_classes, + **kwargs) + return model + + +def swin_base_patch4_window7_224(num_classes: int = 1000, **kwargs): + # trained ImageNet-1K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_base_patch4_window7_224.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=7, + embed_dim=128, + depths=(2, 2, 18, 2), + num_heads=(4, 8, 16, 32), + num_classes=num_classes, + **kwargs) + return model + + +def swin_base_patch4_window12_384(num_classes: int = 1000, **kwargs): + # trained ImageNet-1K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_base_patch4_window12_384.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=12, + embed_dim=128, + depths=(2, 2, 18, 2), + num_heads=(4, 8, 16, 32), + num_classes=num_classes, + **kwargs) + return model + + +def swin_base_patch4_window7_224_in22k(num_classes: int = 21841, **kwargs): + # trained ImageNet-22K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_base_patch4_window7_224_22k.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=7, + embed_dim=128, + depths=(2, 2, 18, 2), + num_heads=(4, 8, 16, 32), + num_classes=num_classes, + **kwargs) + return model + + +def swin_base_patch4_window12_384_in22k(num_classes: int = 21841, **kwargs): + # trained ImageNet-22K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_base_patch4_window12_384_22k.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=12, + embed_dim=128, + depths=(2, 2, 18, 2), + num_heads=(4, 8, 16, 32), + num_classes=num_classes, + **kwargs) + return model + + +def swin_large_patch4_window7_224_in22k(num_classes: int = 21841, **kwargs): + # trained ImageNet-22K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_large_patch4_window7_224_22k.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=7, + embed_dim=192, + depths=(2, 2, 18, 2), + num_heads=(6, 12, 24, 48), + num_classes=num_classes, + **kwargs) + return model + + +def swin_large_patch4_window12_384_in22k(num_classes: int = 21841, **kwargs): + # trained ImageNet-22K + # https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_large_patch4_window12_384_22k.pth + model = SwinTransformer(in_chans=3, + patch_size=4, + window_size=12, + embed_dim=192, + depths=(2, 2, 18, 2), + num_heads=(6, 12, 24, 48), + num_classes=num_classes, + **kwargs) + return model diff --git a/model/backbones/vit_pytorch.py b/model/backbones/vit_pytorch.py index 9ebeb3d39..5f34d2884 100644 --- a/model/backbones/vit_pytorch.py +++ b/model/backbones/vit_pytorch.py @@ -27,8 +27,8 @@ import torch import torch.nn as nn import torch.nn.functional as F -from torch._six import container_abcs - +# from torch._six import container_abcs +import collections.abc as container_abcs # From PyTorch internals def _ntuple(n): @@ -135,6 +135,40 @@ def forward(self, x): x = self.drop(x) return x +class kNNAttention(nn.Module): + def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.,topk=90): + super().__init__() + self.num_heads = num_heads + head_dim = dim // num_heads + # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights + self.scale = qk_scale or head_dim ** -0.5 + + self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) + self.attn_drop = nn.Dropout(attn_drop) + self.proj = nn.Linear(dim, dim) + self.proj_drop = nn.Dropout(proj_drop) + self.topk = topk + + def forward(self, x, islast): + B, N, C = x.shape + qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple) + attn = (q @ k.transpose(-2, -1)) * self.scale + if(islast == False): + # the core code block + mask=torch.zeros(B,self.num_heads,N,N,device=x.device,requires_grad=False) + index=torch.topk(attn,k=self.topk,dim=-1,largest=True)[1] + mask.scatter_(-1,index,1.) + attn=torch.where(mask>0,attn,torch.full_like(attn,float('-inf'))) + # end of the core code block + + attn = attn.softmax(dim=-1) + attn = self.attn_drop(attn) + + x = (attn @ v).transpose(1, 2).reshape(B, N, C) + x = self.proj(x) + x = self.proj_drop(x) + return x, attn class Attention(nn.Module): def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.): @@ -161,7 +195,7 @@ def forward(self, x): x = (attn @ v).transpose(1, 2).reshape(B, N, C) x = self.proj(x) x = self.proj_drop(x) - return x + return x, attn class Block(nn.Module): @@ -170,18 +204,19 @@ def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm): super().__init__() self.norm1 = norm_layer(dim) - self.attn = Attention( - dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop) + self.attn = kNNAttention( + dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop, topk=100) # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() self.norm2 = norm_layer(dim) mlp_hidden_dim = int(dim * mlp_ratio) self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop) - def forward(self, x): - x = x + self.drop_path(self.attn(self.norm1(x))) + def forward(self, x, islast): + atte_output, weight = self.attn(self.norm1(x), islast) + x = x + self.drop_path(atte_output) x = x + self.drop_path(self.mlp(self.norm2(x))) - return x + return x, weight class PatchEmbed(nn.Module): @@ -287,6 +322,19 @@ def forward(self, x): x = x.flatten(2).transpose(1, 2) # [64, 8, 768] return x +class Part_Attention(nn.Module): + def __init__(self): + super(Part_Attention, self).__init__() + + def forward(self, x): + length = len(x) + last_map = x[0] + for i in range(1, length): + last_map = torch.matmul(x[i], last_map) + last_map = last_map[:,:,0,1:] + + _, max_inx = last_map.max(2) + return _, max_inx class TransReID(nn.Module): """ Transformer-based Object Re-Identification @@ -351,6 +399,7 @@ def __init__(self, img_size=224, patch_size=16, stride_size=16, in_chans=3, num_ trunc_normal_(self.pos_embed, std=.02) self.apply(self._init_weights) + self.part_select = Part_Attention() def _init_weights(self, m): if isinstance(m, nn.Linear): @@ -391,17 +440,29 @@ def forward_features(self, x, camera_id, view_id): x = self.pos_drop(x) if self.local_feature: - for blk in self.blocks[:-1]: + for blk in self.blocks: x = blk(x) return x else: - for blk in self.blocks: - x = blk(x) - - x = self.norm(x) - - return x[:, 0] + attn_weights = [] + tokens = [[] for i in range(x.shape[0])] + for blk in self.blocks[:-1]: + x, weights = blk(x, False) + attn_weights.append(weights) + temp_num, temp_inx = self.part_select(attn_weights) + for i in range(B): + tokens[i].extend(x[i, temp_inx[i, :12]]) + + tokens = [torch.stack(token) for token in tokens] + tokens = torch.stack(tokens).squeeze(1) + concat = torch.cat((x[:, 0].unsqueeze(1), tokens), dim=1) + last_blk = self.blocks[-1] + islast = True + last_x, last_weights = last_blk(concat, islast) + last_encoded = self.norm(last_x) + + return last_encoded[:, 0] def forward(self, x, cam_label=None, view_label=None): x = self.forward_features(x, cam_label, view_label) diff --git a/model/make_model.py b/model/make_model.py index 131349201..371030af3 100644 --- a/model/make_model.py +++ b/model/make_model.py @@ -1,5 +1,7 @@ import torch import torch.nn as nn + +from .backbones.model import * from .backbones.resnet import ResNet, Bottleneck import copy from .backbones.vit_pytorch import vit_base_patch16_224_TransReID, vit_small_patch16_224_TransReID, deit_small_patch16_224_TransReID @@ -148,6 +150,12 @@ def __init__(self, num_classes, camera_num, view_num, cfg, factory): if cfg.MODEL.TRANSFORMER_TYPE == 'deit_small_patch16_224_TransReID': self.in_planes = 384 if pretrain_choice == 'imagenet': + # weights_dict = torch.load(model_path, map_location='cuda:0')["model"] + # # 删除有关分类类别的权重 + # for k in list(weights_dict.keys()): + # if "head" in k: + # del weights_dict[k] + # print(self.base.load_state_dict(weights_dict, strict=False)) self.base.load_param(model_path) print('Loading pretrained ImageNet model......from {}'.format(model_path)) @@ -387,7 +395,8 @@ def load_param_finetune(self, model_path): 'vit_base_patch16_224_TransReID': vit_base_patch16_224_TransReID, 'deit_base_patch16_224_TransReID': vit_base_patch16_224_TransReID, 'vit_small_patch16_224_TransReID': vit_small_patch16_224_TransReID, - 'deit_small_patch16_224_TransReID': deit_small_patch16_224_TransReID + 'deit_small_patch16_224_TransReID': deit_small_patch16_224_TransReID, + 'swin': swin_base_patch4_window7_224_in22k } def make_model(cfg, num_class, camera_num, view_num): diff --git a/train.py b/train.py index 2659d63c9..153f107d7 100644 --- a/train.py +++ b/train.py @@ -66,6 +66,7 @@ def set_seed(seed): train_loader, train_loader_normal, val_loader, num_query, num_classes, camera_num, view_num = make_dataloader(cfg) model = make_model(cfg, num_class=num_classes, camera_num=camera_num, view_num = view_num) + print(model) loss_func, center_criterion = make_loss(cfg, num_classes=num_classes)