Evaluation on COCO-test-dev

[kulkarnikeerti]

Hi,

I am trying to test the yolo_nano model with test2017 set. I have already downloaded the test2017 dataset locally. And I tried the exact command mentioned in the readme file, that is

python eval.py -d coco-test \ --cuda \ -m yolov_nano \ --weight path/to/weight \ --img_size 640 \ --root path/to/dataset/

And the readme file says, this generates the .json file. However, unfortunately I end up with the error FileNotFoundError: [Errno 2] No such file or directory: '/data/COCO/annotations/image_info_test-dev2017.json'. This is understandable because there is no such file.

@yjh0410 Did I miss something here? How do I test the model on test dataset?

Thanks!

[yjh0410]

@kulkarnikeerti

You can download it from here: https://github.com/yjh0410/PyTorch_YOLO-Family/releases/download/yolo-weight/image_info_test-dev2017.json

You might also need this, if you want to evaluate the model on COCO2017 test(not test-dev): https://github.com/yjh0410/PyTorch_YOLO-Family/releases/download/yolo-weight/image_info_test2017.json

[kulkarnikeerti]

@yjh0410 Thanks for the links. But I guess they are broken. I don't find anything there. I don't see any releases folder in the repo. Below is the result with the link.

[yjh0410]

@kulkarnikeerti Please try to download from here:

github

gitjub

[kulkarnikeerti]

@yjh0410 Thank you very much.

Just one last clarification: the results of YOLO_Nano in the table contains three different rows. YOLO-Nano-320, YOLO-Nano-416, YOLO-Nano-512.

what does these numbers 320, 416, 512 in the name indicate? Are these the input image size?

[yjh0410]

@kulkarnikeerti Yes. For example, 320 means that the input image is resized to shape 320x320.

[kulkarnikeerti]

Okay. Thanks a lot again @yjh0410