A clear PyTorch template for swift model building.
models/build.py
.solver/loss.py
.data/dataset.py
& data/loader.py
.metrics/eval.py
.solver/solver.py
!scripts/{exp_id}-model_key_config.sh
.+--- .gitignore
+--- archive (generated files & dataset)
| +--- README.md
+--- bin (utility script)
| +--- README.md
| +--- template.py
+--- config.py (options)
+--- data (data fetching related)
| +--- dataset.py
| +--- fetcher.py
| +--- loader.py
| +--- README.md
+--- expr (experiment directory)
+--- main.py (everything start from here)
+--- metrics (metric used)
| +--- eval.py
| +--- fid.py
| +--- README.md
+--- models (model architecture related)
| +--- build.py (the wrapper for models)
| +--- discriminator.py
| +--- generator.py
| +--- layers.py
| +--- mapping_network.py
| +--- README.md
+--- README.md
+--- requirements.txt
+--- scripts (training related shell scripts)
| +--- train.sh
+--- solver (training related)
| +--- loss.py
| +--- misc.py
| +--- solver.py
| +--- utils.py
+--- utils (utility functions)
| +--- checkpoint.py
| +--- file.py
| +--- image.py
| +--- logger.py
| +--- misc.py
| +--- model.py
I referred StarGAN v2's official implementation when crafting this template, so don't be surprised if you find some code is similar.
BTW, if you want to deploy your model, you may want to check out this template.