search

yangsenius / TransPose

PyTorch Implementation for "TransPose: Keypoint localization via Transformer", ICCV 2021.
https://github.com/yangsenius/TransPose/releases/download/paper/transpose.pdf
MIT License
353 stars 56 forks source link

Performance in higher resolutions？like 384*288 #3

Closed leeyegy closed 3 years ago

leeyegy commented 3 years ago

In your paper, only results of 256*192 images are presented. And TransPose-H-A6 performs better than HRNET-32+DRAKPOSE on COCO validation set according to your paper (75.8AP VS 75.6AP).

Hence, I am curious about —— does it still perform better than HRNET if increase the resolutoin?（like 384*288）

it would be appretiated if you can share the experiment results in different resolution.

yangsenius commented 3 years ago

Sorry, @leeyegy, we didn't train TransPose models on 384*288 input resolution.

Because the computational complexity of the self-attention layer is quadratic to the input sequences (HxW), training with 384*288 input resolution will consume much more GPU memory (~5 times than 256x192 input resolution). We may conduct experiments in different resolutions and report results in the future.