From #9, I was wondering if you can provide the 3dpw file or perhaps the preprocessing code?

I followed the SPIN and your 3dhp preprocessing script but the resulted in some PA-MPJPE around 120 using your models on the test npz I put together myself (16 joints). Thanks!

It‘s a great work! And can you upload the The corresponding pre-trained model to evaluate any wild video and to see the visualization result if possible! Best wishes!

Hello. Thanks for your work. I am trying to train a GCN model using the command python3 run_baseline.py --note pretrain --dropout 0 --lr 2e-2 --epochs 100 --posenet_name 'gcn' --checkpoint './checkpoint/pretrain_baseline' --keypoints gt, but an error occurs. Traceback (most recent call last): File "run_baseline.py", line 102, in main(args) File "run_baseline.py", line 65, in main glob_step, args.lr_decay, args.lr_gamma, max_norm=args.max_norm) File "/home/hkuit155/Documents/PoseAug/function_baseline/model_pos_train.py", line 41, in train outputs_3d = model_pos(inputs_2d) File "/home/hkuit155/anaconda3/envs/poseaug/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call result = self.forward(*input, **kwargs) File "/home/hkuit155/Documents/PoseAug/models_baseline/gcn/sem_gcn.py", line 104, in forward out = self.gconv_input(x) File "/home/hkuit155/anaconda3/envs/poseaug/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call result = self.forward(*input, **kwargs) File "/home/hkuit155/anaconda3/envs/poseaug/lib/python3.6/site-packages/torch/nn/modules/container.py", line 92, in forward input = module(input) File "/home/hkuit155/anaconda3/envs/poseaug/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call result = self.forward(*input, **kwargs) File "/home/hkuit155/Documents/PoseAug/models_baseline/gcn/sem_gcn.py", line 28, in forward x = self.gconv(x).transpose(1, 2) File "/home/hkuit155/anaconda3/envs/poseaug/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call result = self.forward(*input, **kwargs) File "/home/hkuit155/Documents/PoseAug/models_baseline/gcn/sem_graph_conv.py", line 43, in forward output = torch.matmul(adj * M, h0) + torch.matmul(adj * (1 - M), h1) RuntimeError: invalid argument 6: wrong matrix size at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:492

A same error also occurs when training ST-GCN

Traceback (most recent call last): File "run_baseline.py", line 102, in main(args) File "run_baseline.py", line 65, in main glob_step, args.lr_decay, args.lr_gamma, max_norm=args.max_norm) File "/home/hkuit155/Documents/PoseAug/function_baseline/model_pos_train.py", line 41, in train outputs_3d = model_pos(inputs_2d) File "/home/hkuit155/anaconda3/envs/poseaug/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call result = self.forward(*input, **kwargs) File "/home/hkuit155/Documents/PoseAug/models_baseline/models_st_gcn/st_gcn_single_frame_test.py", line 461, in forward x = torch.matmul(x, C) # nx2x17 RuntimeError: invalid argument 6: wrong matrix size at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:492

Hi nice work and thanks for releasing your code.

could you please point me to what data for MPI you use to train and validate on? Also, can you tell me what file contains your skeleton and MPI specific dataset class?

Also, do you only evaluate with MpJpE or do you use PcK amd any other metrics on MPI?

Hi. I am running your code aiming at lifting the 2d pose to 3d. However, in the pre-processing step, I found that camera parameters are in need. Can I just use the models to lift the 2d pose to the 3d pose without camera calibration? For example, I obtained an image with humans whose camera parameters are unknown, and using pose lifting model after obtaining the 2d pose.

Can this model run a single picture? When I ran a picture, there was some misalignment. The model used was MLP trained by PoseAug, and I found that the result was not very satisfactory.

Your work is great! Thanks. when I run the command python3 run_baseline.py --note pretrain --lr 1e-3 --posenet_name 'videopose' --checkpoint './checkpoint/pretrain_baseline' --keypoints gt, the following error was generated Can i trouble you to check what is wrong? Thanks!

Hello,

Sorry to bother you again, may I know which joint is not needed for your 2D input? Is it head, or nose? I understand that your input and output are both 16 joints, and nose is not predicted by the model.

Many thanks, Shiyang

I am assuming your training data is H36M poses not the meshes and you are evaluating on pose on MPI-INF-3DHP as well, without using SMPL parameters from both. Further, you apply poseAug on 2D and 3D poses of H36M to get training pairs. At that time you would use 16 joint skeleton model of

So, my questions are how do you evaluate given that the skeleton model of MPI-INF-3DHP is 23 joints models and some basic joints have no correspondence from H36M to MPI-INF-3DHP, e.g. Thorex, head_top, Spine are not there in H36M, however, H36M have Neck and Head joint.

How did you evaluate MPI-INF-3DHP in a cross-dataset-testing-scenrio given skeleton model discrepancy?

I guess you only use a subset of joints to evaluate?

Hi author. Thanks for the great work!

I'm still not clear on data pre-processing. Could you please release the code or point me how to use additional in-the-wild 2D poses (MPII & LSP) to train the 2D discriminator? Which you mention in paper page 6 "Results on 3DHP (cross-scenario) part".

Thank you for your excellent research.

https://github.com/jfzhang95/PoseAug/blob/main/data_extra/bone_length_npy/hm36s15678_bl_templates.npy

I wonder how this file was created.

Hello! I want to ask some questions about PosDiscriminator. First, get_ BoneVecbypose3d and get_ Pose3dbyBoneVec can be extended to 20 key points, right? Second, local KCS uses indexes, such as hb_ Idx=[0, 3, 4, 5, 6]. This is the index of the joint points included in the right leg. I went to see the bone map and found that it does not match. Third, PosDiscriminator inputs 3D posture and outputs the precise posture. Fourth, what are the rules for the human body to split the five parts. Look forward to your reply, thank you! @Garfield-kh

I am trying to use this training frame to train a baseline trajectory model (predict root node camera space position). It works but loss seems weird: the test MPJPE is always around 110.

Is there anything I miss? Have you tried train any trajectory model for PoseAug model?