residual-flows
code for "Residual Flows for Invertible Generative Modeling".
MIT License
Stars
261
Ecosystems:
Python
Each checkpoint file contains:
- Model params at end of training.
- EMA params accumulated throughout training.
- Arguments used for training the model.
- Test bits/dim of the model with EMA params.
Use the --resume [checkpt.pth] argument to evaluate, fine-tune, or sample from the model.
The CelebA-HQ 256x256 file was split into mutiple chunks using the method from here. Download all three chunks and use the following command to extract the full checkpoint file:
cat celebahq256_resflow_16-16-16-16-16-16.pth.split.gz_* | gunzip -c > celebahq256_resflow_16-16-16-16-16-16.pth