VIBE
VIBE copied to clipboard
mkocabas
Official implementation of CVPR2020 paper "VIBE: Video Inference for Human Body Pose and Shape Estimation"
Thanks for your work! I'm trying to run your Google Collaboratory example. With the following code: # Install the other requirements !pip install torch==1.4.0 numpy==1.17.5 !pip install git+https://github.com/giacaglia/pytube.git --upgrade !pip...
thanks a lot for such a great repo How could I set inference just for one body and ignore others? (for the largest boundary that Yolo detects.)
Hi, THX for sharing your work. I met some questions when reproduce your best model. When I visualize the output mesh, the best model provided by yours fit well but...
Could you please tell me how to use blender or keyshot with your camera parameters? I want to get rendered images with high quality.
As far as I currently understand it, the x,y,z coordinates of every joint is displayed in the output file, but the proper rotations of said joints don't seem to be...
Hello! I was wondering if you might clarify which data VIBE uses to generate (1) the SotA results referenced in the paper and (2) the pretrained weights used in the...
threedpw_utils.py 采用默认的threedpw_utils代码处理3dpw的数据时候，采用下面的代码获得vertices #output = smpl(betas=shape, body_pose=pose[:,3:], global_orient=pose[:,:3], transl=trans)#原始代码 发现这个vertices与spin网络获得cam并不一致，原因是这里传入了3dpw的trans参数， 而vibe模型参数的中，网络训练的输出，并没有传入trans参数，代码如下，不知道是不是弄错了， 或者在代码里面哪里做了转换，望指教，谢谢 pred_output = self.smpl( betas=pred_shape, body_pose=pred_rotmat[:, 1:], global_orient=pred_rotmat[:, 0].unsqueeze(1), pose2rot=False )
Dear doctor: 1. my operating system and the version: OSX mojave 2. python 3.7 3. no gpu After i install packeage needed, i run the expamle code: python demo.py --vid_file...
Hello, I noticed that during Inference the input to the model is not split into chunks of seqlen=16. All the frames of the video are feed at once. In Inference.py...
