Skip to content

Latest commit

 

History

History
243 lines (186 loc) · 19.3 KB

README_SwinIR.md

File metadata and controls

243 lines (186 loc) · 19.3 KB
Raw

Made some changes to train (large) swinIR models for x1 (only denoising), x2, x3, and x4 scenarios more efficiently

Following are some noteable changes:

1. Fp16 support
2. Tensorboard support
3. Absolute paths are required in the config (json) files
4. Added bash scripts for distributed training on single and multiple nodes
5. Infer and save images during trainig for `save_test_out` under `datasets` in a config for quality evaluation of the intermediate checkpoints
6. Added bash scripts for distributed training on single (`bash_util_scripts/train_swinIR_dist.sh`) and multiple (`bash_util_scripts/train_swinIR_dist_multinode.sh`) nodes

Instructions for distributed training:

1. Update following arguments in the `bash_util_scripts/train_swinIR_dist.sh` file:
    * `num_gpus_per_node` argument
    * set `opt` to either:
      - options/swinir/train_swinir_large_sr_realworld_x2_psnr_fp16.json for PSNR (phase1) trainings or
      - options/swinir/train_swinir_large_sr_realworld_x2_gan_fp16.json for GAN (phase2) trainings

2. Launch training:
    - For single node training, run the following cmd: `bash train_swinIR_dist.sh 1 0 127.0.0.1`
    - For multi node training on Nvidia GPUs:
        1. Create /job/.ssh/config file with information of all the nodes in the following format:
            - An e.g. of 2 nodes system where ps-0 is the node from where is the training job is launched
              ```
              Host ps-0
                HostName 192.168.0.38
                Port 42076
                User foo
                StrictHostKeyChecking no
                UserKnownHostsFile /dev/null

              Host worker-0
                HostName 192.168.0.38
                Port 48575
                User foo
                StrictHostKeyChecking no
                UserKnownHostsFile /dev/null

              Host worker-1
                HostName 192.168.0.101
                Port 46577
                User foo
                StrictHostKeyChecking no
                UserKnownHostsFile /dev/null
              ```
        2. Run the cmd to kick off the training: `bash train_swinIR_dist_multinode.sh /home/saghotra/git/KAIR/bash_util_scripts/train_swinIR_dist.sh /tmp`
          - You can see errors in /tmp/DistributedJobLogs
  - For multinode training on AMD, use _AMD versions of the bash script after running `create_hostfile_AMD.sh`



SwinIR: Image Restoration Using Shifted Window Transformer

paper | supplementary | visual results | original project page | online Colab demo

arXiv GitHub Stars download google colab logo

Image restoration is a long-standing low-level vision problem that aims to restore high-quality images from low-quality images (e.g., downscaled, noisy and compressed images). While state-of-the-art image restoration methods are based on convolutional neural networks, few attempts have been made with Transformers which show impressive performance on high-level vision tasks. In this paper, we propose a strong baseline model SwinIR for image restoration based on the Swin Transformer. SwinIR consists of three parts: shallow feature extraction, deep feature extraction and high-quality image reconstruction. In particular, the deep feature extraction module is composed of several residual Swin Transformer blocks (RSTB), each of which has several Swin Transformer layers together with a residual connection. We conduct experiments on three representative tasks: image super-resolution (including classical, lightweight and real-world image super-resolution), image denoising (including grayscale and color image denoising) and JPEG compression artifact reduction. Experimental results demonstrate that SwinIR outperforms state-of-the-art methods on different tasks by up to 0.14~0.45dB, while the total number of parameters can be reduced by up to 67%.

Dataset Preparation

Training and testing sets can be downloaded as follows. Please put them in trainsets and testsets respectively.

Task Training Set Testing Set
classical/lightweight image SR DIV2K (800 training images) or DIV2K +Flickr2K (2650 images) set5 + Set14 + BSD100 + Urban100 + Manga109 download all
real-world image SR SwinIR-M (middle size): DIV2K (800 training images) +Flickr2K (2650 images) + OST (10324 images, sky,water,grass,mountain,building,plant,animal)
SwinIR-L (large size): DIV2K + Flickr2K + OST + WED(4744 images) + FFHQ (first 2000 images, face) + Manga109 (manga) + SCUT-CTW1500 (first 100 training images, texts)

*We use the first practical degradation model BSRGAN, ICCV2021 GitHub Stars for real-world image SR		 RealSRSet+5images
color/grayscale image denoising DIV2K (800 training images) + Flickr2K (2650 images) + BSD500 (400 training&testing images) + WED(4744 images) grayscale: Set12 + BSD68 + Urban100
color: CBSD68 + Kodak24 + McMaster + Urban100 download all
JPEG compression artifact reduction DIV2K (800 training images) + Flickr2K (2650 images) + BSD500 (400 training&testing images) + WED(4744 images) grayscale: Classic5 +LIVE1 download all

Training

To train SwinIR, run the following commands. You may need to change the dataroot_H, dataroot_L, scale factor, noisel level, JPEG level, G_optimizer_lr, G_scheduler_milestones, etc. in the json file for different settings.

# 001 Classical Image SR (middle size)
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_sr_classical.json  --dist True

# 002 Lightweight Image SR (small size)
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_sr_lightweight.json  --dist True

# 003 Real-World Image SR (middle size)
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_sr_realworld_psnr.json  --dist True
# before training gan, put the PSNR-oriented model into superresolution/swinir_sr_realworld_x4_gan/models/
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_sr_realworld_gan.json  --dist True

# 004 Grayscale Image Deoising (middle size)
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_denoising_gray.json  --dist True

# 005 Color Image Deoising (middle size)
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_denoising_color.json  --dist True

# 006 JPEG Compression Artifact Reduction (middle size)
python -m torch.distributed.launch --nproc_per_node=8 --master_port=1234 main_train_psnr.py --opt options/swinir/train_swinir_car_jpeg.json  --dist True

You can also train above models using DataParallel as follows, but it will be slower.

# 001 Classical Image SR (middle size)
python main_train_psnr.py --opt options/swinir/train_swinir_sr_classical.json

...

Note:

1, We fine-tune X3/X4/X8 (or noise=25/50, or JPEG=10/20/30) models from the X2 (or noise=15, or JPEG=40) model, so that total_iteration can be halved to save training time. In this case, we halve the initial learning rate and lr_milestones accordingly. This way has similar performance as training from scratch.

2, For SR, we use different kinds of Upsampler in classical/lightweight/real-world image SR for the purpose of fair comparison with existing works.

3, We did not re-train the models after cleaning the codes. Feel free to open an issue if you meet any problems.

Testing

Following command will download the pretrained models and put them in model_zoo/swinir. All visual results of SwinIR can be downloaded here.

If you are too lazy to prepare the datasets, please follow the guide in the original project page, where you can start testing in a minute. We also provide an online Colab demo for real-world image SR google colab logo for comparison with the first practical degradation model BSRGAN (ICCV2021) GitHub Stars and a recent model RealESRGAN. Try to test your own images on Colab!

# 001 Classical Image Super-Resolution (middle size)
# Note that --training_patch_size is just used to differentiate two different settings in Table 2 of the paper. Images are NOT tested patch by patch.
# (setting1: when model is trained on DIV2K and with training_patch_size=48)
python main_test_swinir.py --task classical_sr --scale 2 --training_patch_size 48 --model_path model_zoo/swinir/001_classicalSR_DIV2K_s48w8_SwinIR-M_x2.pth --folder_lq testsets/set5/LR_bicubic/X2 --folder_gt testsets/set5/HR
python main_test_swinir.py --task classical_sr --scale 3 --training_patch_size 48 --model_path model_zoo/swinir/001_classicalSR_DIV2K_s48w8_SwinIR-M_x3.pth --folder_lq testsets/set5/LR_bicubic/X3 --folder_gt testsets/set5/HR
python main_test_swinir.py --task classical_sr --scale 4 --training_patch_size 48 --model_path model_zoo/swinir/001_classicalSR_DIV2K_s48w8_SwinIR-M_x4.pth --folder_lq testsets/set5/LR_bicubic/X4 --folder_gt testsets/set5/HR
python main_test_swinir.py --task classical_sr --scale 8 --training_patch_size 48 --model_path model_zoo/swinir/001_classicalSR_DIV2K_s48w8_SwinIR-M_x8.pth --folder_lq testsets/set5/LR_bicubic/X8 --folder_gt testsets/set5/HR

# (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
python main_test_swinir.py --task classical_sr --scale 2 --training_patch_size 64 --model_path model_zoo/swinir/001_classicalSR_DF2K_s64w8_SwinIR-M_x2.pth --folder_lq testsets/set5/LR_bicubic/X2 --folder_gt testsets/set5/HR
python main_test_swinir.py --task classical_sr --scale 3 --training_patch_size 64 --model_path model_zoo/swinir/001_classicalSR_DF2K_s64w8_SwinIR-M_x3.pth --folder_lq testsets/set5/LR_bicubic/X3 --folder_gt testsets/set5/HR
python main_test_swinir.py --task classical_sr --scale 4 --training_patch_size 64 --model_path model_zoo/swinir/001_classicalSR_DF2K_s64w8_SwinIR-M_x4.pth --folder_lq testsets/set5/LR_bicubic/X4 --folder_gt testsets/set5/HR
python main_test_swinir.py --task classical_sr --scale 8 --training_patch_size 64 --model_path model_zoo/swinir/001_classicalSR_DF2K_s64w8_SwinIR-M_x8.pth --folder_lq testsets/set5/LR_bicubic/X8 --folder_gt testsets/set5/HR


# 002 Lightweight Image Super-Resolution (small size)
python main_test_swinir.py --task lightweight_sr --scale 2 --model_path model_zoo/swinir/002_lightweightSR_DIV2K_s64w8_SwinIR-S_x2.pth --folder_lq testsets/set5/LR_bicubic/X2 --folder_gt testsets/set5/HR
python main_test_swinir.py --task lightweight_sr --scale 3 --model_path model_zoo/swinir/002_lightweightSR_DIV2K_s64w8_SwinIR-S_x3.pth --folder_lq testsets/set5/LR_bicubic/X3 --folder_gt testsets/set5/HR
python main_test_swinir.py --task lightweight_sr --scale 4 --model_path model_zoo/swinir/002_lightweightSR_DIV2K_s64w8_SwinIR-S_x4.pth --folder_lq testsets/set5/LR_bicubic/X4 --folder_gt testsets/set5/HR


# 003 Real-World Image Super-Resolution (use --tile 400 if you run out-of-memory)
# (middle size)
python main_test_swinir.py --task real_sr --scale 4 --model_path model_zoo/swinir/003_realSR_BSRGAN_DFO_s64w8_SwinIR-M_x4_GAN.pth --folder_lq testsets/RealSRSet+5images

# (larger size + trained on more datasets)
python main_test_swinir.py --task real_sr --scale 4 --large_model --model_path model_zoo/swinir/003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN.pth --folder_lq testsets/RealSRSet+5images


# 004 Grayscale Image Deoising (middle size)
python main_test_swinir.py --task gray_dn --noise 15 --model_path model_zoo/swinir/004_grayDN_DFWB_s128w8_SwinIR-M_noise15.pth --folder_gt testsets/set12
python main_test_swinir.py --task gray_dn --noise 25 --model_path model_zoo/swinir/004_grayDN_DFWB_s128w8_SwinIR-M_noise25.pth --folder_gt testsets/set12
python main_test_swinir.py --task gray_dn --noise 50 --model_path model_zoo/swinir/004_grayDN_DFWB_s128w8_SwinIR-M_noise50.pth --folder_gt testsets/set12


# 005 Color Image Deoising (middle size)
python main_test_swinir.py --task color_dn --noise 15 --model_path model_zoo/swinir/005_colorDN_DFWB_s128w8_SwinIR-M_noise15.pth --folder_gt testsets/McMaster
python main_test_swinir.py --task color_dn --noise 25 --model_path model_zoo/swinir/005_colorDN_DFWB_s128w8_SwinIR-M_noise25.pth --folder_gt testsets/McMaster
python main_test_swinir.py --task color_dn --noise 50 --model_path model_zoo/swinir/005_colorDN_DFWB_s128w8_SwinIR-M_noise50.pth --folder_gt testsets/McMaster


# 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding uses 8x8 blocks)
python main_test_swinir.py --task jpeg_car --jpeg 10 --model_path model_zoo/swinir/006_CAR_DFWB_s126w7_SwinIR-M_jpeg10.pth --folder_gt testsets/classic5
python main_test_swinir.py --task jpeg_car --jpeg 20 --model_path model_zoo/swinir/006_CAR_DFWB_s126w7_SwinIR-M_jpeg20.pth --folder_gt testsets/classic5
python main_test_swinir.py --task jpeg_car --jpeg 30 --model_path model_zoo/swinir/006_CAR_DFWB_s126w7_SwinIR-M_jpeg30.pth --folder_gt testsets/classic5
python main_test_swinir.py --task jpeg_car --jpeg 40 --model_path model_zoo/swinir/006_CAR_DFWB_s126w7_SwinIR-M_jpeg40.pth --folder_gt testsets/classic5

Results

Classical Image Super-Resolution (click me)

Lightweight Image Super-Resolution

Real-World Image Super-Resolution

    Real-World Image (x4) BSRGAN, ICCV2021 Real-ESRGAN SwinIR (ours)
Grayscale Image Deoising

Color Image Deoising

JPEG Compression Artifact Reduction

Please refer to the paper and the original project page for more results.

Citation

@article{liang2021swinir,
    title={SwinIR: Image Restoration Using Swin Transformer},
    author={Liang, Jingyun and Cao, Jiezhang and Sun, Guolei and Zhang, Kai and Van Gool, Luc and Timofte, Radu},
    journal={arXiv preprint arXiv:2108.10257},
    year={2021}
}