JeffLi

Hi @dldaisy, sorry for the misleading. The two config files are for two different models (a 29 joints model and a 24 joints model). The two models are trained and...

In our experiments, the 24 joints model is sightly better.

Hi @SecretMG, you may need to manually comment the code about `DistributedDataParallel` and `DistributedSampler`.

Hi @dldaisy, In the case of partial observation, the current HybrIK model cannot handle it correctly. Since the joint rotations are calculated based on the detected keypoints, when the keypoint...

Thanks for your suggestion, we will update the code later.

Hi @seoha-kim , You can refer to [L222](https://github.com/Jeff-sjtu/HybrIK/blob/main/hybrik/datasets/h36m_smpl.py#L222), [L224](https://github.com/Jeff-sjtu/HybrIK/blob/main/hybrik/datasets/h36m_smpl.py#L224) and [L246](https://github.com/Jeff-sjtu/HybrIK/blob/main/hybrik/datasets/h36m_smpl.py#L246). As long as the annotation files provide `bbox`, `cam_param` and `root_coord`, I think the dataloader can do the rest...

Yes, that's what I mean. Here is our manuscript code. There might be some variable names that should be modified for compatibility with the current code. ``` python import copy...

Hi @wanzysky , I think you are correct. This pull request is for improving the stability of the transformation. The original version will fail in some corner cases. I am...

Hi @wanzysky , We use the batch operation in the latest version. I think the speed would be faster. I remember I also needed more than 30 minutes per epoch...

Actually, you just need to run 140 epochs. The extra epoch is adopting `dpg` from our project AlphaPose.