AttributeError: type object 'CTLModel' has no attribute '_load_model_state'

[real-wind]

Thank you for the hardwork and sharing of your code, I have encountered some issues while attempting to replicate the results for the PRCC dataset with the CTL model.
I executed the inference script:
python /GEFF/Scripts/inference.py prcc CTL --reid_config /GEFF/ReIDModules/centroids_reid/configs/256_resnet50.yml --dataset_path /GEFF/datasets/prcc --device "cuda:0" --reid_checkpoint GEFF/checkpoints/CTL/dukemtmcreid_resnet50_256_128_epoch_120.ckpt --alpha 0.75 --detection_threshold 0.7
Actual Results:
Setting seed to: 0
Traceback (most recent call last):
File "/GEFF/Scripts/inference.py", line 472, in
main(args)
File "/GEFF/Scripts/inference.py", line 467, in main
run_inference(args, config=config)
File "/GEFF/Scripts/inference.py", line 399, in run_inference
qd_pids, qd_camids, qd_clothes_ids, g_pids, g_camids, g_clothes_ids = reid_inference_on_prcc(args, config)
File "/GEFF/Scripts/inference.py", line 184, in reid_inference_on_prcc
reid_model.init_model(config, args.reid_checkpoint)
File "/GEFF/./ReIDModules/CTL_model.py", line 17, in init_model
self.model = CTL._load_model_state(checkpoint)
AttributeError: type object 'CTLModel' has no attribute '_load_model_state'

I have double-checked the steps and parameters, but I still face difficulties .Any guidance on how to proceed or troubleshoot this issue would be greatly appreciated.

[darkushin]

Hi @real-wind,

The issue you encountered seems to be related to the environment and not an issue with the code itself (more specifically, the function that fails is the pytorch-lightning function _load_model_state as the original CTL model was built using pytorch-lightning).

Setting up the environment for CTL was quite challenging, therefore, we don't support it in the Colab notebook demo, and I would recommend using one of the other models we support (CAL/AIM).

If you still want to use the CTL model I would suggest to first try to run the model as described by the authors of the model (centroids-reid repo), and once you are able to use their model, try to modify the CTL init_model function in our repo so that it will be able to load the pre-trained checkpoint of the model.

Good luck!