search

microsoft / Deep3DFaceReconstruction

Accurate 3D Face Reconstruction with Weakly-Supervised Learning: From Single Image to Image Set (CVPRW 2019)
MIT License
2.15k stars 438 forks source link

Why using ResNet-50 to regress the target vector instead of ResNet-101？ #159

Open HhansssW opened 3 years ago

HhansssW commented 3 years ago

Hi! Could you please give me some information about the reason choosing ResNet-50 ? Since I saw from the 3DMM-CNN that they used ResNet-101 to regress 198 3DMM parameters which is less than that in your work (257 parameters). Thank you so much.

YuDeng commented 3 years ago

For the first question, we did not use ResNet-101 because it is kind of heavy and its performance is only slightly better compare to ResNet-50 in our experiment.

For the second question, we follow MoFA to set the dimension of 3DMM parameters in order to compare with it fairly.

HhansssW commented 3 years ago

For the first question, we did not use ResNet-101 because it is kind of heavy and its performance is only slightly better compare to ResNet-50 in our experiment.

For the second question, we follow MoFA to set the dimension of 3DMM parameters in order to compare with it fairly.

Thanks ! ;)