训练自己的数据集问题

[meituan]

训练自己的数据集问题

[ExWang]

When generate_coco_format_labels, this repo make img_id by osp.basename(img_path).split('.')[0] in datasets.py. You will get str label.

BUT, when evaluating results, this repo make img_id by int(path.stem) if path.stem.isnumeric() else path.stem in convert_to_coco_format in evaler.py. You may get int label.

Then, coco tools will check these img_id, so lead this AssertionError.

So I suggest to modify image_id = int(path.stem) if path.stem.isnumeric() else path.stem to image_id = path.stem to solve this error.

[mingwz]

Thanks for your reply! I solve the problem by this way. @ExWang

[nidetaoge]

Have already pulled the newest commit, but still get the same error:
pred = anno.loadRes(pred_json)
File "/opt/conda/lib/python3.8/site-packages/pycocotools/coco.py", line 341, in loadRes
assert set(annsImgIds) == (set(annsImgIds) & set(self.getImgIds())),
AssertionError: Results do not correspond to current coco set

[mingwz]

老哥你解决了吗？我也是遇到了相同的问题

[ruoyuchen0903]

so how can I solve it? 2022-06-27 15:05:30"Nebula Wang" ***@***.***>写道： when generate_coco_format_labels， this repo make img_id by osp.basename(img_path).split('.')[0] in datasets.py. You will get 'str' label. BUT, when evaluating results, this repo make img_id by 'int(path.stem) if path.stem.isnumeric() else path.stem'. You may get 'int' label. After, coco tools will check these img_id, so lead this AssertionError — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ExWang]

so how can I solve it? 2022-06-27 15:05:30"Nebula Wang" @.>写道： when generate_coco_format_labels， this repo make img_id by osp.basename(img_path).split('.')[0] in datasets.py. You will get 'str' label. BUT, when evaluating results, this repo make img_id by 'int(path.stem) if path.stem.isnumeric() else path.stem'. You may get 'int' label. After, coco tools will check these img_id, so lead this AssertionError — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.>

[moxigk]

之前的错误是：
TypeError: '<' not supported between instances of 'str' and 'int'

我根据你的方法改了，出现了这个问题
AssertionError: Results do not correspond to current coco set

是不是数据集的名字必须是数字吗？不能包含英文吗或者其他字符吗？

[xiaoluxiaoluxiaolu]

Hello, I would like to know which file needs to be modified

[ExWang]

Hello, I would like to know which file needs to be modified

no file need to be modified now, some one fix it, pull new repo will slove the problem.

[ExWang]

之前的错误是： TypeError: '<' not supported between instances of 'str' and 'int'

我根据你的方法改了，出现了这个问题 AssertionError: Results do not correspond to current coco set

是不是数据集的名字必须是数字吗？不能包含英文吗或者其他字符吗？

我看不是必须数字的，最早这个repo生成image_id用的 img_id = osp.basename(img_path).split('.')[0] 是str的id。但是在后面predict的时候，用的 img_id = int(path.stem) if path.stem.isnumeric() else path.stem ,也就是说，假设你的图片名字是数字的，就会转成int的id。这就造成了coco检查img_id的时候，发现他们对不上，引发了这个Error.

不过这个不统一的问题好像在最近的一个commit中修复了，pull一下最新的就行

[Mobu59]

@nidetaoge have you fixed the problem?

[yutao007]

将这里的images_id修改后，仍然还是报错。感觉这一块有问题没有完全解决。其中我的环境是：
torch 1.10.0+cu111
torchaudio 0.10.0+rocm4.1
torchvision 0.11.0+cu111

修改处
def save_one_json(predn, jdict, path, class_map):
# Save one JSON result {"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}
#image_id = int(path.stem) if path.stem.isnumeric() else path.stem
image_id = path.stem

报错处
assert set(annsImgIds) == (set(annsImgIds) & set(self.getImgIds())),
AssertionError: Results do not correspond to current coco set
terminate called without an active exception
Aborted (core dumped)

[qybing]

1.把datasets.py文件的img_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem
2.把evaler.py文件的image_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem
就可以运行起来了

[Mobu59]

1.把datasets.py文件的img_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem 2.把evaler.py文件的image_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem 就可以运行起来了

Thanks, I made it

[yutao007]

谢谢博主解答，我已经按照提示将datasets.py evaler.py val.py 三个文件中关于image_id的赋值都改成了path.stem
但还是报错，如果别人没错的话，那应该就是我这边数据集或者别的问题了
assert set(annsImgIds) == (set(annsImgIds) & set(self.getImgIds())),
AssertionError: Results do not correspond to current coco set

[yutao007]

问题解决了，要记得删掉先前跑数据生成的json文件。

[luckyeap2020]

之前的错误是： TypeError: '<' not supported between instances of 'str' and 'int'

我根据你的方法改了，出现了这个问题 AssertionError: Results do not correspond to current coco set

是不是数据集的名字必须是数字吗？不能包含英文吗或者其他字符吗？

不用改源代码，只用一个粗暴的方式解决了：

1. 数据集中文件的名称只用数字
2. 删除之前训练产生的所有过程文件，即只保留原始数据集的图片和标注

[ruabliuqiu]

训练前，自己数据集分布和制作方法跟yolov5一样，不需要json格式标签。我参考这个博主的方法训练成功了，https://blog.csdn.net/DeepLearning_/article/details/125511165?spm=1001.2014.3001.5502

[shenshi19]

训练前，自己数据集分布和制作方法跟yolov5一样，不需要json格式标签。我参考这个博主的方法训练成功了，https://blog.csdn.net/DeepLearning_/article/details/125511165?spm=1001.2014.3001.5502
兄弟，你的yaml文件是什么样子的，我参考了你发的链接还是有问题的

[Yaomingyang]

官方已经解决这个问题了，重新下载工程，按照docs/Train_custom_data.md运行即可

[ruabliuqiu]

训练前，自己数据集分布和制作方法跟yolov5一样，不需要json格式标签。我参考这个博主的方法训练成功了，https://blog.csdn.net/DeepLearning_/article/details/125511165?spm=1001.2014.3001.5502
兄弟，你的yaml文件是什么样子的，我参考了你发的链接还是有问题的

yaml中只保留train和val，去掉test和anno_path

[yaos666]

1.把datasets.py文件的img_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem 2.把evaler.py文件的image_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem 就可以运行起来了

成功啦，必须把datasets.py、evaler.py都改了才能运行起来，只改evaler.py文件还是会报错的哟

[shenshi19]

1.把datasets.py文件的img_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem 2.把evaler.py文件的image_id = int(path.stem) if path.stem.isnumeric() else path.stem改成img_id = path.stem 就可以运行起来了

成功啦，必须把datasets.py、evaler.py都改了才能运行起来，只改evaler.py文件还是会报错的哟

新版的好像datasets.py好像img_id-int()那里改了

[suzauddola]

Evaluating mAP by pycocotools.
Saving runs/train/ttop_6s_3004/predictions.json...
ERROR in evaluate and save model.
ERROR in training loop or eval/save model.

Training completed in 0.039 hours.
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
Loading and preparing results...
Traceback (most recent call last):
File "tools/train.py", line 94, in
main(args)
File "tools/train.py", line 84, in main
trainer.train()
File "/YOLOv6/yolov6/core/engine.py", line 70, in train
self.train_in_loop()
File "/YOLOv6/yolov6/core/engine.py", line 89, in train_in_loop
self.eval_and_save()
File "/YOLOv6/yolov6/core/engine.py", line 115, in eval_and_save
self.eval_model()
File "/YOLOv6/yolov6/core/engine.py", line 134, in eval_model
results = eval.run(self.data_dict,
File "/yolov6/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "/YOLOv6/tools/eval.py", line 83, in run
eval_result = val.eval_model(pred_result, model, dataloader, task)
File "/YOLOv6/yolov6/core/evaler.py", line 127, in eval_model
pred = anno.loadRes(pred_json)
File "/home/suza/.conda/envs/yolov6/lib/python3.8/site-packages/pycocotools/coco.py", line 327, in loadRes
assert set(annsImgIds) == (set(annsImgIds) & set(self.getImgIds())),
AssertionError: Results do not correspond to current coco set

When generate_coco_format_labels, this repo make img_id by osp.basename(img_path).split('.')[0] in datasets.py. You will get str label.

BUT, when evaluating results, this repo make img_id by int(path.stem) if path.stem.isnumeric() else path.stem in convert_to_coco_format in evaler.py. You may get int label.

Then, coco tools will check these img_id, so lead this AssertionError.

So I suggest to modify image_id = int(path.stem) if path.stem.isnumeric() else path.stem to image_id = path.stem to solve this error.

After changing to "image_id = path.stem" now this error appears !!

[jcrdeep]

#207 (comment)

This resolved in my case