cifr-pytorch (This project is in progress.)

Continuous Implicit Feature Representation

What's different from LIIF

Adversarial Training (Generator: Encoder + LIIF, Discriminator: U-Net based)
An Encoder is StyleGAN based architecture (Noise Injection for generating fine-grained details)
Contextual Loss

0. Installation

apt install ninja-build
pip install -r requirements.txt

1. Dataset

Download Train Data (HR images) and Validation Data (HR images) from DIV2K Dataset.

curl -L http://data.vision.ee.ethz.ch/cvl/DIV2K/DIV2K_train_HR.zip -o DIV2K_train_HR.zip
curl -L http://data.vision.ee.ethz.ch/cvl/DIV2K/DIV2K_valid_HR.zip -o DIV2K_valid_HR.zip

Decompress thease files to ./sr_dataset folder.

unzip DIV2K_train_HR.zip
unzip DIV2K_valid_HR.zip

You have the folder structure.

./sr_dataset
    |
    |-- DIV2K_train_HR
    |
    |-- DIV2K_valid_HR

2. Train

For single-gpu training,

export PYTHONPATH=$(pwd)
python tools/train.py --config configs/div2k_stylegan_sn_liif.py

For multi-gpu training,

python -m torch.distributed.run --nproc_per_node=4 \
tools/train_dist.py \
--config configs/div2k_stylegan_sn_liif.py

3. Inference

export PYTHONPATH=$(pwd)
python tools/inference_liif.py \
--config configs/div2k_stylegan_sn_liif.py \
--ckpt work_dir/div2k_stylegan_sn_liif/checkpoints/000200.pth \
--img test.jpg

References

Really thank the authors of LIIF for sharing their codes and research.

[1] Learning Continuous Image Representation with Local Implicit Image Function, Yinbo Chen et al.

shi510/cifr-pytorch