Color Transfer with Differentiable Histogram

This project is based on Differentiable Histogram with Hard-Binning.

Prerequisites

• Python 3.6.x
• PyTorch 1.6.0 & torchvision 0.7.0

Dataset

Dataset was generated from edges2shoes dataset using create_dataset.py

python create_dataset.py --original_dir "edges2shoes" --new_dir "edges2shoes_triplets" --n_repeat 1

Training

Train model

python main.py --mode train --data_dir [data_directory] --out_dir [output_directory] --n_epoch 100 --resize 143 --crop 128 --batch_size 30 --hist_loss mae --lr_decay_start 30 --lr_decay_n 70

Resume training from checkpoint

python main.py --mode train --data_dir [data_directory] --out_dir [output_directory] --n_epoch 100 --resize 143 --crop 128 --batch_size 30 --hist_loss mae --lr_decay_start 30 --lr_decay_n 70 --pretrain_path ./[output_directory]/xxx/xxx.pt

Plot loss stats from train.json

python plot.py --dir [output_directory]

It will look for train.json in the directory and output plots as result.png.

See more options available

python main.py -h

Testing

python main.py --mode test --crop 128 --resize 143 --data_dir datasets/edges2shoes --pretrain_path [output_directory]/xxx.pt

This generates all images from test set and save them to ./checkpoints/xxx/images/test/.