About NEUDET training error

[MrDec]

[02/20 09:59:19 detectron2]: Full config saved to output/vitb/neu-det_1shot/config.yaml
('NEUDET_test',)
('NEUDET_test',)
Traceback (most recent call last):
File "tools/train_net.py", line 208, in
launch(
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/engine/launch.py", line 67, in launch
mp.spawn(
File "/home/dww/anaconda3/envs/detectron/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 240, in spawn
return start_processes(fn, args, nprocs, join, daemon, start_method='spawn')
File "/home/dww/anaconda3/envs/detectron/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 198, in start_processes
while not context.join():
File "/home/dww/anaconda3/envs/detectron/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 160, in join
raise ProcessRaisedException(msg, error_index, failed_process.pid)
torch.multiprocessing.spawn.ProcessRaisedException:

-- Process 1 terminated with the following error:
Traceback (most recent call last):
File "/home/dww/anaconda3/envs/detectron/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap
fn(i, *args)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/engine/launch.py", line 125, in _distributed_worker
main_func(*args)
File "/home/dww/OD/CDFSOD-benchmark/tools/train_net.py", line 196, in main
trainer = Trainer(cfg)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/engine/defaults.py", line 381, in init
model = self.build_model(cfg)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/engine/defaults.py", line 521, in build_model
model = build_model(cfg)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/modeling/meta_arch/build.py", line 22, in build_model
model = META_ARCH_REGISTRY.get(meta_arch)(cfg)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/config/config.py", line 174, in wrapped
explicit_args = _get_args_from_config(from_config_func, *args, **kwargs)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/config/config.py", line 229, in _get_args_from_config
ret = from_config_func(*args, **kwargs)
File "/home/dww/OD/CDFSOD-benchmark/tools/../detectron2/modeling/meta_arch/devit.py", line 909, in from_config
"seen_cids": SEEN_CLS_DICT[cfg.DATASETS.TRAIN[0]],
KeyError: 'NEUDET_1shot'

[MrDec]

您好，非常有趣的工作，目前有两个问题。

1. 我尝试在NEUDET数据集上运行，结果报错。不知道是哪里的原因。
2. 这个方法在生成原型的过程中只适用于一张图像上只有一个目标的情况吗？如果我想在一张图像上有多个目标的情形下是否可以适用？
   期待您的回答。

[ohMargin]

1.NEUDET数据集运行时的报错可以查看lib/categories.py中数据集名称部分，可以把datasets_name和CLASS_NAME里的NEUDET都修改成NEU-DET，应该能解决你的问题。
2.生成prototype的时候主要面向的问题都是单个目标的，但最终也会出现测试时有多个目标的情况，感兴趣的话可以再做一些实验试试

[lovelyqian]

@MrDec we just released the 1st CD-FSOD challenge under NTIRE CVPR25 workshop, may we invite you to join this as participant. More info could be find here: https://codalab.lisn.upsaclay.fr/competitions/21851

[ohMargin]

请参考代码./lib/categories.py的73行，这里原先注释掉了后续数据集，datasets_name = ('ArTaxOr', 'clipart1k')#, 'DIOR', 'FISH', 'NEUDET', 'UODD')。可能是这个问题影响了你的代码运行

[Blizzard-stack]

@ohMargin 你好，我按您说的把datasets_name和CLASS_NAME里的NEUDET都修改成NEU-DET，但运行还是显示找不到，训练时显示'DIOR', 'FISH', 'NEUDET', 'UODD'的1shot、5shot、10shot都找不到，想请问一下您这个报错如何解决，谢谢

[Blizzard-stack]

@MrDec 你好，我也遇到了这个问题，请问一下您解决了嘛，您是如何解决的？

[ohMargin]

@Blizzard-stack 请参考代码文件夹路径lib/categories.py的73行，这里原先注释掉了后续数据集，datasets_name = ('ArTaxOr', 'clipart1k')#, 'DIOR', 'FISH', 'NEUDET', 'UODD')。 把这里的#去掉了吗？

[Blizzard-stack]

@ohMargin你好，我已经成功复现，运行结果与论文中表格1中的结果差不多，我想请问一下如何得到图4中的可视化结果，谢谢

[ohMargin]

@Blizzard-stack 图4有很多可视化，想问你具体需要的是哪部分的可视化方式呢？

[Blizzard-stack]

你好，我需要的是实现图4中(b),(c),(e)的可视化，谢谢 Chengzhenhao ***@***.***

…

---Original--- From: "Xingyu ***@***.***> Date: Sat, Mar 29, 2025 16:11 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [lovelyqian/CDFSOD-benchmark] About NEUDET training error (Issue#36) @Blizzard-stack 图4有很多可视化，想问你具体需要的是哪部分的可视化方式呢？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***> ohMargin left a comment (lovelyqian/CDFSOD-benchmark#36) @Blizzard-stack 图4有很多可视化，想问你具体需要的是哪部分的可视化方式呢？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[ohMargin]

4(b)是对提取prototypes后对其做的特征余弦相似度，作图简单易实现。
4(c)是对实例做重新权重划分后，对于不同权重的图片整理后依据从高到低的顺序挑选出一些图片呈现。
4(e)是对域特征做得TSNE提取，需要存储特征后进行TSNE工具的划分。

[Blizzard-stack]

你好，可以提供一下它们可视化代码嘛，谢谢 Chengzhenhao ***@***.***

…

---Original--- From: "Xingyu ***@***.***> Date: Sun, Mar 30, 2025 11:11 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [lovelyqian/CDFSOD-benchmark] About NEUDET training error (Issue#36) 4(b)是对提取prototypes后对其做的特征余弦相似度，作图简单易实现。 4(c)是对实例做重新权重划分后，对于不同权重的图片整理后依据从高到低的顺序挑选出一些图片呈现。 4(e)是对域特征做得TSNE提取，需要存储特征后进行TSNE工具的划分。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***> ohMargin left a comment (lovelyqian/CDFSOD-benchmark#36) 4(b)是对提取prototypes后对其做的特征余弦相似度，作图简单易实现。 4(c)是对实例做重新权重划分后，对于不同权重的图片整理后依据从高到低的顺序挑选出一些图片呈现。 4(e)是对域特征做得TSNE提取，需要存储特征后进行TSNE工具的划分。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>