search

mks0601 / I2L-MeshNet_RELEASE

Official PyTorch implementation of "I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image", ECCV 2020
MIT License
720 stars 128 forks source link

Run on hands #27

Closed FranciscoGomez90 closed 3 years ago

FranciscoGomez90 commented 3 years ago

I am having trouble when running MeshNet + RootNet on hands. I downloaded the rootnet model, from this repo and cloned rootnet. When I run the rootnet demo with the model provided here, an error is shown RuntimeError: Error(s) in loading state_dict for DataParallel: Missing key(s) in state_dict: "module.root.deconv_layers.0.weight", "module.root.deconv_layers.1.weight",.... How can I run RootNet with the trained model you provide in this repo? thank you.

mks0601 commented 3 years ago

Hmm.. I haven't provided pre-trained RootNet in this repo? You should use RootNet repo in here.

FranciscoGomez90 commented 3 years ago

Yes, but in the readme there is a part that states

FreiHAND dataset Download I2L-MeshNet trained on [FreiHand]. Download bounding boxs and root joint coordinates (from RootNet) of [FreiHAND]. bbox is from Detectron2.

The Download I2L-MeshNet trained on [FreiHand] link (https://drive.google.com/drive/folders/1NGEScDf2tYtrloLGinqV97SRvfpSJ4o7) gives you two models: i2L-mesh and RootNet. Are these models the correct ones for hand pose estimation? thanks

mks0601 commented 3 years ago

Oh I forgot this. The RootNet in the drive is trained using my modified code so input normalization and the module names are slightly different. You should rename module.root_net keys in the snapshot file to module.root keys. Also, just normalize 0-255 image to 0-1 instead of using this line

FranciscoGomez90 commented 3 years ago

Could you please provide the complete rootnet you used?

mks0601 commented 3 years ago

3DMPPE-RootNet.zip

I guess this would work, but not perfectly sure. Sorry I'm too busy to check this :(