search

postech-ami / FastMETRO

[ECCV'22] Official PyTorch Implementation of "Cross-Attention of Disentangled Modalities for 3D Human Mesh Recovery with Transformers"
https://fastmetro.github.io/
MIT License
164 stars 14 forks source link

About pretrained CNN network #14

Closed Hao-hui closed 1 year ago

Hao-hui commented 1 year ago

In PARE(ICCV2021) and a number of other articles, they mentioned that training ResNet or HR-Net with attitude estimation task on COCO dataset can help improve the performance of human reconstruction network. However, it seems that in your experiment, the network was pre-trained with image-net. Have you done any relevant experiments？

FastMETRO commented 1 year ago

Hello,

For the fair comparison with METRO, we used the same backbone initialized with the same pre-trained weights (i.e., pre-trained with ImageNet). We haven’t conducted experiments on the different backbone initialization with other pre-trained weights (e.g., pre-trained with COCO).

Thanks for your interest in our work!!

FastMETRO commented 1 year ago

Please reopen this issue if you need more help regarding this.