search

cherubicXN / logocap

Offical code of LOGO-CAP (CVPR' 22). https://arxiv.org/abs/2109.03622
33 stars 7 forks source link

Can't find the .yaml of OCHuman #3

Closed zhexiangrenjian closed 1 year ago

zhexiangrenjian commented 1 year ago

Hi, Thank you for the amazing work! Could you please release the .YAML files of OCHuman？

cherubicXN commented 1 year ago

Actually, I did not create a YAML file for OCHuman as we did not do the training on this dataset and the format is the same as COCO. So, as long as you downloaded (and uncompressed) the OCHuman dataset in the data directory with the name ochuman, you can make the inference by 

python tools/test.py --cfg experiments/logocap-hrnet-w32-coco.yaml \
 --ckpt weights/logocap/logocap-hrnet-w32-coco.pth.tar  \
DATASET.TEST {split} DATASET.ROOT data/ochuman

where {split} is "eval" for the testing split and "val" for the validation split.

It should be noted that I replaced the split name of testing from "test" to "eval" to avoid a conflict with the COCO test-dev dataset that doesn't have ground truth for offline evaluation.

Changing the following lines https://github.com/cherubicXN/logocap/blob/39d104fdaf4970bab91392fe81897e450daaffb9/experiments/logocap-hrnet-w32-coco.yaml#L27-L28 also works.