wjn922 / ReferFormer

[CVPR2022] Official Implementation of ReferFormer
Apache License 2.0
322 stars 25 forks source link

Joint Training Settings #34

Closed lszw11 closed 1 year ago

lszw11 commented 1 year ago

Thank you for your great work! I met a problem when I reproduced the result of your joint train. Do any weights of the backbone network need to be loaded during the joint training e.g. video-swin-b-kinetics400-22k. I observed that there is no option to load any weights in the published joint training script. Thanks!

wjn922 commented 1 year ago

Thanks for your interest.

Yes, you need to load the imagenet-pretrained backbone weights for joint training. Please download the swin/video-swin backbone weights from their official repos first, and then use the arg --backbone_pretrained to specify the weight path. https://github.com/wjn922/ReferFormer/blob/9c8f237adc260c512a1c5ecfc7aee81b8282649a/opts.py#L33

lszw11 commented 1 year ago

Thank you for your reply!