Frequently Asked Questions

1. Can I Use conda Instead Of mamba?

Yes. In the setup instructions, you can replace mamba with conda in the environment commands.

For example:

conda env create -f env.yml
conda activate gf5
conda env update -f env.yml --prune

mamba is a drop-in replacement for conda with faster environment resolution. The resulting gf5 environment should be the same.

2. Can I Keep My Custom Code In My Own Git Repository?

Yes. You can set up two Git remotes for your local repository:

• one remote for pulling updates from the teaching repository
• one remote for pushing your own custom code to your personal repository

For example:

git remote rename origin teaching
git remote add myrepo git@github.com:YOUR_USERNAME/YOUR_REPO.git

Then pull course updates from the teaching repository:

git pull teaching main

and push your own work to your repository:

git push myrepo main

Use your actual repository URL in place of git@github.com:YOUR_USERNAME/YOUR_REPO.git.

3. Which SMPL Model Should I Download?

See Part 2: SMPL Model Setup for the detailed download instructions. The short version is: use the standard SMPL body model download, unzip it, and place the extracted smpl/ folder inside assets/ so the model files appear under:

assets/smpl/models/

4. Why Does SMPL Not Load On Windows?

If the blocky/proxy assets load but the real SMPL asset does not, and the Use LBS checkbox stays unavailable, check your PyTorch and NumPy versions. This Windows failure is tracked in GitHub issue #1.

The reported broken environment mixed torch 2.10.0+cpu with numpy 2.4.4. That can produce PyTorch DLL loading errors and SMPLX-to-NumPy conversion errors. One confirmed fix is:

conda activate gf5
python -m pip uninstall -y torch torchvision torchaudio
python -m pip install torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cpu
python -m pip install numpy==1.26.4 --force-reinstall

Then test:

python -c "import torch; print(torch.__version__)"
python -c "import numpy; print(numpy.__version__)"
python -c "import smplx; print('smplx ok')"

Expected versions for this workaround are torch 2.3.1+cpu and numpy 1.26.4.

If Windows also reports an OpenMP duplicate-runtime warning, run:

set KMP_DUPLICATE_LIB_OK=TRUE
python viewer\asset_viewer.py --smpl-model assets\smpl\models\SMPL_MALE.pkl

5. Why Does The Character Reconstruction Link Fail On Some WiFi Networks?

If the character reconstruction link from the internal Part 3 page works on one network but fails on another with an error such as DNS_PROBE_FINISHED_NXDOMAIN, the most likely cause is DNS caching or filtering on that network. This can happen when a subdomain was created recently: one DNS resolver may know about it, while another may temporarily remember an older "does not exist" response.

This is not usually a problem with the GF5 code or the reconstruction server.

Temporary fixes:

• try a different network, such as a mobile hotspot
• in Chrome, enable Secure DNS with Cloudflare or Google: Settings -> Privacy and security -> Security -> Use secure DNS
• wait and try again later, since DNS caches usually expire automatically

If you can reach the page on mobile data but not on WiFi, report the WiFi network name and the exact browser error message.

6. What If The HY-Motion Demo Says I Have Hit A Hugging Face Limit?

The public HY-Motion demo runs on Hugging Face, so anonymous usage can sometimes hit a rate or quota limit. If this happens, sign up for a free Hugging Face account and log in before trying the demo again. Free accounts usually receive a higher quota than anonymous users.