Installation instructions need to be updated (currently dont work) and missing items

[Alex-Kupin]

The installation instructions do not work for inference.

To get it to work a few things need to be done. I am also on linux for those wondering.

git clone --recursive https://github.com/Pixel-Talk/PEAR.git
cd PEAR

conda create -n pear python=3.9.22
conda activate pear

# Make sure cuda version is 11.8
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
# Remove torch==2.0.1+cu118, torchvision, torchaudio, lightning from the requirements.txt
pip install -r requirements.txt
pip install "git+https://github.com/facebookresearch/pytorch3d.git" --no-build-isolation
pip install chumpy --no-build-isolation
pip install lightning==2.0.9 pytorch-lightning==2.0.9

It also took some looking to find the correct SMPL models you can find them here:
SMPL_NEUTRAL.pkl
SMPLX_NEUTRAL_2020.npz
FLAME_2020
You will also need smpl_mean_params.npz

It is missing a lbs_map_smplx_512.npz in the assets/SMPLX directory. Cannot proceed without it.

[WuJH2001]

Hi @Alex-Kupin ! Thank you very much for pointing out our mistake. We have updated the file accordingly. We have also provided the complete asset.zip at this url. If it is only for inference, you can also download them from our Hugging Face project page

[Alex-Kupin]

Thank you! I love your user profile pic.

What is also missing is the smplx_tex.mtl and Flame_masks.pkl (i found these online). The assets zip was very helpful and had everything inside.

I am stuck now getting it to run
🎯 Running...
🎯 Running EHM pipeline...
❌ Error in launch_viz: Not compiled with GPU support

What CUDA version/toolkit are you using?

[WuJH2001]

Hi @Alex-Kupin , We’re very glad to hear that you were able to run it successfully. Regarding CUDA, we have verified that it works properly with versions 11.8, 12.2, and 12.4.