ggsonic / hidt

Official repository for the paper High-Resolution Daytime Translation Without Domain Labels (CVPR2020, Oral)

High-Resolution Daytime Translation Without Domain Labels

Project Page | Video Explanation | Paper | Appendix | TwoMinutePapers

Official PyTorch implementation (only inference part) for the paper I. Anokhin, P. Solovev, D. Korzhenkov, A. Kharlamov, T. Khakhulin, A. Silvestrov, S. Nikolenko, V. Lempitsky, and G. Sterkin. "High-Resolution Daytime Translation Without Domain Labels." In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

Installation

Make sure that you use python >= 3.7. We have tested it with conda package manager. If you are new to conda, proceed to https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html

conda create -n hidt python=3.7
conda activate hidt

Clone the repo

git clone https://github.com/saic-mdal/HiDT.git

Install requirenments

cd HiDT
pip install -r requirements.txt

Inference

Daytime translation, upsampling with G_enh

CUDA_VISIBLE_DEVICES=0 PYTHONPATH=$(pwd):${PYTHONPATH} \
python ./bin/infer_on_folders.py \
    --content-dir ./images/daytime/content/ \
    --style-dir ./images/daytime/styles/ \
    --cfg-path ./configs/daytime.yaml \
    --chk-path ./trained_models/generator/daytime.pt \
    --enh-path ./trained_models/enhancer/enhancer.pth \
    --enhancement generator

Daytime translation, generator in fully convolutional mode, no postprocessing

CUDA_VISIBLE_DEVICES=0 PYTHONPATH=$(pwd):${PYTHONPATH} \
python ./bin/infer_on_folders.py \
    --content-dir ./images/daytime/content/ \
    --style-dir ./images/daytime/styles/ \
    --cfg-path ./configs/daytime.yaml \
    --chk-path ./trained_models/generator/daytime.pt \
    --enhancement fullconv

Model, trained on wikiart, upsampling with G_enh

CUDA_VISIBLE_DEVICES=0 PYTHONPATH=$(pwd):${PYTHONPATH} \
python ./bin/infer_on_folders.py \
    --content-dir ./images/wikiart/content/ \
    --style-dir ./images/wikiart/styles/ \
    --cfg-path ./configs/wikiart.yaml \
    --chk-path ./trained_models/generator/wikiart.pt \
    --enh-path ./trained_models/enhancer/enhancer.pth \
    --enhancement generator

Model, trained on wikiart, generator in fully convolutional mode, no postprocessing

CUDA_VISIBLE_DEVICES=0 PYTHONPATH=$(pwd):${PYTHONPATH} \
python ./bin/infer_on_folders.py \
    --content-dir ./images/wikiart/content/ \
    --style-dir ./images/wikiart/styles/ \
    --cfg-path ./configs/wikiart.yaml \
    --chk-path ./trained_models/generator/wikiart.pt \
    --enhancement fullconv

Citation

If you found our work useful, please don't forget to cite

@inproceedings{Anokhin_2020_CVPR,
  author = {Anokhin, Ivan and
            Solovev, Pavel and
            Korzhenkov, Denis and
            Kharlamov, Alexey and
            Khakhulin, Taras and
            Silvestrov, Alexey and
            Nikolenko, Sergey and
            Lempitsky, Victor and
            Sterkin, Gleb
  },
  title = {High-Resolution Daytime Translation Without Domain Labels},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {June},
  year = {2020},
}