Unofficial implementation of DualPatchNorm using tensorflow2. Include ablation code too.
Paper: Dual PatchNorm(Google Research and Brain Team)
Not completed codes.
git clone git@github.com:dslisleedh/DualPatchNorm-tensorflow2.git
cd DualPatchNorm-tensorflow2
conda env create -f environment.yaml
conda activate dpn
python train.py
If you have never downloaded the imagenet_resized 64x64 using tensorflow_datasets, run donwload_datasets.py before run train.py
Default config is set run HPO that takes lots of time.
Disable HPO if you just want to run code for once.
python train.py hpo=False hparams.$ANY_HPARAMS_IN_CONFIG=...