search

bytedance / coconut_cvpr2024

Apache License 2.0
148 stars 6 forks source link

where is the annotation? #3

Closed z-jiaming closed 6 months ago

z-jiaming commented 6 months ago

Thanks for your good work!

I can`t find the segmentation annotations in your files.

For each anno in 'annotations', instances_train2017.json of COCO provides them like:

Snipaste_2024-04-25_17-06-53

But it doesn't seem to be the same in your file.

image

Do I use the wrong way?

z-jiaming commented 6 months ago

Thus, I can not use pycocotools.coco.COCO directly.

image
xdeng7 commented 6 months ago

Thus, I can not use pycocotools.coco.COCO directly.

image

we only provide the panoptic mask in the download link, it looks like you are using instance masks with pycocotools, we recommend you to follow the tutorial under the folder "kmaxdeeplab_instance" in this repo to convert the panoptic masks to instance masks. The tool will just ignore the "stuff" masks but generate "thing" instance masks that will can be used in pycocotools.

z-jiaming commented 6 months ago

Thanks a lot for your reply! I try to use the tool but find some bugs. I fix it, you can see the changes in https://github.com/bytedance/coconut_cvpr2024/pull/4