Yolov5: The exported label txt seems have problem

[wwdok]

Hi, @glenn-jocher i manage to use python detect.py --save-txt to semi-auto label images, but when i set Open Dir and Change Save Dir in labelImg，the labelImg can not display the exported bbox, and its command line window appears error:

Traceback (most recent call last):
  File "<string>", line 1268, in openNextImg
  File "<string>", line 1035, in loadFile
  File "<string>", line 1427, in loadYOLOTXTByFilename
  File "Z:\home\darrenl\tmp\labelImg\build-tools\build\labelImg\out00-PYZ.pyz\libs.yolo_io", line 112, in __init__
  File "Z:\home\darrenl\tmp\labelImg\build-tools\build\labelImg\out00-PYZ.pyz\libs.yolo_io", line 142, in parseYoloFormat
ValueError: too many values to unpack

If i set Change Save Dir to another empty folder, it will not occur error, so i doubt it is the problem of exported label txt, could you have a try ?

[wwdok]

@glenn-jocher, i found the reason ! Because the exported txt has a whitespace at the end of each line, if you delete this whitespace, everything is normal !

[wwdok]

Just change the code to f.write((('%g ' * len(line)).rstrip() + '\n') % line) solved ~

[glenn-jocher]

@wwdok ah I see, very interesting, thank you for raising this issue. Your fix looks good. Would you like to submit a PR for this update?

[wwdok]

@glenn-jocher Yeah, it is my first time to submit a PR, so i must carefuly learn how to do it, i will do it tomorrow😄

[carloslopezjoya]

Hello @wwdok , How to get bounding box coordinates in YOLO object detection and store them in variable?

[wwdok]

@carloslopezjoya You can use python detect.py --save-txt to generate a txt file, inside it, it contains the class id, four coordinates fraction, just like the one you prepared for training. If you want to get coordinates in python codes, they are named xyxy, you can search them in detect.py.

[carloslopezjoya]

@wwdok thanks for your answer, In my case I am running it in an IDE 'spyder' I added return xyxy and it does not return anything, it does not even save the txt
These are my configurations

[carloslopezjoya]

@wwdok In itself I need to store bounding box coordinates in YOLO object detection and store them in a variable

[glenn-jocher]

@carloslopezjoya python detect.py --save-txt saves labels.

[carloslopezjoya]

@glenn-jocher I know, in my case I use this python detect.py --weights best.pt --img 416 --conf 0.4 --source test\images --save-txt and it works, I just need to run it from my IDE and I can't store xyxy in a variable and load in ram

[carloslopezjoya]

@glenn-jocher I am creating an object counter, if I need to store the xyxy coordinates in a variable to manipulate them at the same time

[wwdok]

@carloslopezjoya you could try out (torch.tensor(xyxy).view(1, 4)).tolist()[0] and print out to verify it.

[wwdok]

Hi, @glenn-jocher , I found a place that could be improved, that is adding assert out != source, 'Please set --save-dir to be different from --source ' , if someone set them the same, then the program will delete all the images in the source folder, because the name of the image to be generated is the same as the name in the source folder, if he/her didn't backup the images, that will be an accident ~

[glenn-jocher]

@wwdok thanks for the warning! The recent PR below from a few days ago created a new common saving structure across all the 3 main files (train, test, detect). The repo will now not overwrite an existing directory unless --exist-ok is passed as an argument. If the project/name directory already exists it will be incremented to project/name2 automatically.

• Unified '/project/name' results saving #1377

[glenn-jocher]

@wwdok in the case of detect.py, --output is now --project/--name

[wwdok]

@glenn-jocher Glad to see the file structure becomes more organized and robust !

[carloslopezjoya]

@carloslopezjoya , podrías probar (torch.tensor(xyxy).view(1, 4)).tolist()[0]e imprimir para verificarlo.

thanks @wwdok

[glenn-jocher]

Removing TODO, original issue resolved by PR.