Raphaël

Solving environment: failed ResolvePackageNotFound:

10

when I run "conda env create -f pyskl.yaml", I meet the error as follows: `Collecting package metadata (repodata.json): done Solving environment: failed ResolvePackageNotFound: - mkl_fft==1.3.1=py37hd3c417c_0 - tk==8.6.10=hbc83047_0 - libunistring==0.9.10=h27cfd23_0 -...

How many joints in SMPL-H model ?

1

I find there are 73 joints in SMPL_H model using your code. However, I find [this](https://github.com/nghorbani/human_body_prior) contains 52 joints in SMPL_H model. Who is wrong? Thanks.

I wonder why you translate whole sequence skeleton to make spin joint always be the origin. [here](https://github.com/lshiwjx/2s-AGCN/blob/953c14fc10883cd869646328f5d522e9e9282063/data_gen/preprocess.py#L37) Did you have seen the improvement of such preprocess? I think it might...

In thailand language, whsiper-large-v3 could get word level timestamps, but whisperx could not! Why?

1

Using whisperx, I got final result after alignemnet: `{'segments': [{'start': 1.254, 'end': 6.66, 'text': 'มียุทธศาสตร์ในการศึกษาให้รัฐสภา มู่สู่ความเป็นรัฐสภาที่ทันสมัย', 'words': [{'word': 'มียุทธศาสตร์ในการศึกษาให้รัฐสภา', 'start': 1.254, 'end': 4.137, 'score': 0.292}, {'word': 'มู่สู่ความเป็นรัฐสภาที่ทันสมัย', 'start': 4.217, 'end':...

Everytime I install this repo by pip, I would see the torch 2.x is downloading and installing even if I remove `torch>2` in the `requirements.txt` . I think it is...

I mean that we could use an animation file in new skeleton with same topology and different offsets as the style example.

would torch.float32 type influence precision as you use RNN in gesture decoder?

1

I found that [it](https://github.com/ubisoft/ubisoft-laforge-ZeroEGGS/blob/4992d9ab5e28bc4bb13219a598906562f5786448/ZEGGS/modules.py#L118) would cause that the reuslts might be different when the rnn decoder run many times. I guess this is beacuse of float32 type would harm the...