How to convert VisDrone label format to yolo

[ziyaakhan]

Hi how convert VisDrone annotations to yolo v7 format ?

[frankykubo]

Hello, did you manage to do this?

[ziyaakhan]

Hi, i couldn't do it but i found this https://www.kaggle.com/datasets/mainhatnam/visdrone-2019-mot-train?select=images

[frankykubo]

Thats nice! I am gonna give it a try. Could you fine train any yolo model with that?

Also, I found this:
https://github.com/ultralytics/yolov5/blob/master/data/VisDrone.yaml
which is script to actually transform VisDrone dataset to yolo-awaited shape. But I cant get it to work because of error:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 6.00 GiB total capacity; 5.32 GiB already allocated; 0 bytes free; 5.35 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

which happens after few entries in first epoch.

[ziyaakhan]

Thats nice! I am gonna give it a try. Could you fine train any yolo model with that?

Yes it is compatible with all versions

Also, I found this: https://github.com/ultralytics/yolov5/blob/master/data/VisDrone.yaml which is script to actually transform VisDrone dataset to yolo-awaited shape. But I cant get it to work because of error:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 6.00 GiB total capacity; 5.32 GiB already allocated; 0 bytes free; 5.35 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

which happens after few entries in first epoch.

Vram is full, my suggestion choose a lower batch size

[adityatandon]

If you're looking for a better way to convert the annotations with a cleaner script, I had been working on this sometime back.

I've added the code to convert the annotations as well as the annotations in the YOLO format on my Github here. Feel free to use that for your experiments if you'd like.