image super-resolution¶
Summary¶
Number of checkpoints: 37
Number of configs: 41
Number of papers: 11
ALGORITHM: 11
SwinIR (ICCVW’2021)¶
Task: Image Super-Resolution, Image denoising, JPEG compression artifact reduction
Abstract¶
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%.
Results and models¶
Classical Image Super-Resolution¶
Evaluated on Y channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
| Model | Dataset | Task | Scale | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|---|---|
| swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k | Set5 | Image Super-Resolution | x2 | 38.3240 | 0.9626 | 8 | model | log |
| swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k | Set14 | Image Super-Resolution | x2 | 34.1174 | 0.9230 | 8 | model | log |
| swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k | DIV2K | Image Super-Resolution | x2 | 37.8921 | 0.9481 | 8 | model | log |
| swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k | Set5 | Image Super-Resolution | x3 | 34.8640 | 0.9317 | 8 | model | log |
| swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k | Set14 | Image Super-Resolution | x3 | 30.7669 | 0.8508 | 8 | model | log |
| swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k | DIV2K | Image Super-Resolution | x3 | 34.1397 | 0.8917 | 8 | model | log |
| swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k | Set5 | Image Super-Resolution | x4 | 32.7315 | 0.9029 | 8 | model | log |
| swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k | Set14 | Image Super-Resolution | x4 | 28.9065 | 0.7915 | 8 | model | log |
| swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k | DIV2K | Image Super-Resolution | x4 | 32.0953 | 0.8418 | 8 | model | log |
| swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k | Set5 | Image Super-Resolution | x2 | 38.3971 | 0.9629 | 8 | model | log |
| swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k | Set14 | Image Super-Resolution | x2 | 34.4149 | 0.9252 | 8 | model | log |
| swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k | DIV2K | Image Super-Resolution | x2 | 37.9473 | 0.9488 | 8 | model | log |
| swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k | Set5 | Image Super-Resolution | x3 | 34.9335 | 0.9323 | 8 | model | log |
| swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k | Set14 | Image Super-Resolution | x3 | 30.9258 | 0.8540 | 8 | model | log |
| swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k | DIV2K | Image Super-Resolution | x3 | 34.2830 | 0.8939 | 8 | model | log |
| swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k | Set5 | Image Super-Resolution | x4 | 32.9214 | 0.9053 | 8 | model | log |
| swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k | Set14 | Image Super-Resolution | x4 | 29.0792 | 0.7953 | 8 | model | log |
| swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k | DIV2K | Image Super-Resolution | x4 | 32.3021 | 0.8451 | 8 | model | log |
Lightweight Image Super-Resolution¶
Evaluated on Y channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
| Model | Dataset | Task | Scale | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|---|---|
| swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k | Set5 | Image Super-Resolution | x2 | 38.1289 | 0.9617 | 8 | model | log |
| swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k | Set14 | Image Super-Resolution | x2 | 33.8404 | 0.9207 | 8 | model | log |
| swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k | DIV2K | Image Super-Resolution | x2 | 37.5844 | 0.9459 | 8 | model | log |
| swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k | Set5 | Image Super-Resolution | x3 | 34.6037 | 0.9293 | 8 | model | log |
| swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k | Set14 | Image Super-Resolution | x3 | 30.5340 | 0.8468 | 8 | model | log |
| swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k | DIV2K | Image Super-Resolution | x3 | 33.8394 | 0.8867 | 8 | model | log |
| swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k | Set5 | Image Super-Resolution | x4 | 32.4343 | 0.8984 | 8 | model | log |
| swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k | Set14 | Image Super-Resolution | x4 | 28.7441 | 0.7861 | 8 | model | log |
| swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k | DIV2K | Image Super-Resolution | x4 | 31.8636 | 0.8353 | 8 | model | log |
Real-World Image Super-Resolution¶
Evaluated on Y channels.
The metrics are NIQE .
| Model | Dataset | Task | NIQE | Training Resources | Download |
|---|---|---|---|---|---|
| swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost | RealSRSet+5images | Image Super-Resolution | 5.7975 | 8 | model | log |
| swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost | RealSRSet+5images | Image Super-Resolution | 7.2738 | 8 | model | log |
| swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost | RealSRSet+5images | Image Super-Resolution | 5.2329 | 8 | model | log |
| swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost | RealSRSet+5images | Image Super-Resolution | 7.7460 | 8 | model | log |
| swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost | RealSRSet+5images | Image Super-Resolution | 5.1464 | 8 | model | log |
| swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost | RealSRSet+5images | Image Super-Resolution | 7.6378 | 8 | model | log |
Grayscale Image Deoising¶
Evaluated on grayscale images.
The metrics are PSNR .
| Model | Dataset | Task | PSNR | Training Resources | Download |
|---|---|---|---|---|---|
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15 | Set12 | Image denoising | 33.9731 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15 | BSD68 | Image denoising | 32.5203 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15 | Urban100 | Image denoising | 34.3424 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25 | Set12 | Image denoising | 31.6434 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25 | BSD68 | Image denoising | 30.1377 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25 | Urban100 | Image denoising | 31.9493 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50 | Set12 | Image denoising | 28.5651 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50 | BSD68 | Image denoising | 27.3157 | 8 | model | log |
| swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50 | Urban100 | Image denoising | 28.6626 | 8 | model | log |
Color Image Deoising¶
Evaluated on RGB channels.
The metrics are PSNR .
JPEG Compression Artifact Reduction (grayscale)¶
Evaluated on grayscale images. The metrics are `PSNR / SSIM
| Model | Dataset | Task | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|---|
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10 | Classic5 | JPEG compression artifact reduction | 30.2746 | 0.8254 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10 | LIVE1 | JPEG compression artifact reduction | 29.8611 | 0.8292 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20 | Classic5 | JPEG compression artifact reduction | 32.5331 | 0.8753 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20 | LIVE1 | JPEG compression artifact reduction | 32.2667 | 0.8914 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30 | Classic5 | JPEG compression artifact reduction | 33.7504 | 0.8966 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30 | LIVE1 | JPEG compression artifact reduction | 33.7001 | 0.9179 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40 | Classic5 | JPEG compression artifact reduction | 34.5377 | 0.9087 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40 | LIVE1 | JPEG compression artifact reduction | 34.6846 | 0.9322 | 8 | model | log |
JPEG Compression Artifact Reduction (color)¶
Evaluated on RGB channels.
The metrics are PSNR / SSIM .
| Model | Dataset | Task | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|---|
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10 | Classic5 | JPEG compression artifact reduction | 30.1019 | 0.8217 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10 | LIVE1 | JPEG compression artifact reduction | 28.0676 | 0.8094 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20 | Classic5 | JPEG compression artifact reduction | 32.3489 | 0.8727 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20 | LIVE1 | JPEG compression artifact reduction | 30.4514 | 0.8745 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30 | Classic5 | JPEG compression artifact reduction | 33.6028 | 0.8949 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30 | LIVE1 | JPEG compression artifact reduction | 31.8235 | 0.9023 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40 | Classic5 | JPEG compression artifact reduction | 34.4344 | 0.9076 | 8 | model | log |
| swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40 | LIVE1 | JPEG compression artifact reduction | 32.7610 | 0.9179 | 8 | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
## 001 Classical Image Super-Resolution (middle size)
## (setting1: when model is trained on DIV2K and with training_patch_size=48)
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k.py
## (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k.py
## 002 Lightweight Image Super-Resolution (small size)
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k.py
## 003 Real-World Image Super-Resolution
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py
## 004 Grayscale Image Deoising (middle size)
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50.py
## 005 Color Image Deoising (middle size)
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50.py
## 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding uses 8x8 blocks)
## grayscale
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40.py
## color
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30.py
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40.py
## single-gpu train
## 001 Classical Image Super-Resolution (middle size)
## (setting1: when model is trained on DIV2K and with training_patch_size=48)
python tools/train.py configs/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k.py
python tools/train.py configs/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k.py
python tools/train.py configs/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k.py
## (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
python tools/train.py configs/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k.py
python tools/train.py configs/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k.py
python tools/train.py configs/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k.py
## 002 Lightweight Image Super-Resolution (small size)
python tools/train.py configs/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k.py
python tools/train.py configs/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k.py
python tools/train.py configs/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k.py
## 003 Real-World Image Super-Resolution
python tools/train.py configs/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
python tools/train.py configs/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
python tools/train.py configs/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
python tools/train.py configs/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py
python tools/train.py configs/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py
python tools/train.py configs/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py
## 004 Grayscale Image Deoising (middle size)
python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15.py
python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25.py
python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50.py
## 005 Color Image Deoising (middle size)
python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15.py
python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25.py
python tools/train.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50.py
## 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding uses 8x8 blocks)
## grayscale
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10.py
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20.py
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30.py
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40.py
## color
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10.py
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20.py
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30.py
python tools/train.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40.py
## multi-gpu train
## 001 Classical Image Super-Resolution (middle size)
## (setting1: when model is trained on DIV2K and with training_patch_size=48)
./tools/dist_train.sh configs/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k.py 8
./tools/dist_train.sh configs/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k.py 8
./tools/dist_train.sh configs/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k.py 8
## (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
./tools/dist_train.sh configs/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k.py 8
./tools/dist_train.sh configs/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k.py 8
./tools/dist_train.sh configs/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k.py 8
## 002 Lightweight Image Super-Resolution (small size)
./tools/dist_train.sh configs/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k.py 8
./tools/dist_train.sh configs/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k.py 8
./tools/dist_train.sh configs/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k.py 8
## 003 Real-World Image Super-Resolution
./tools/dist_train.sh configs/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py 8
./tools/dist_train.sh configs/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py 8
./tools/dist_train.sh configs/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py 8
./tools/dist_train.sh configs/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py 8
./tools/dist_train.sh configs/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py 8
./tools/dist_train.sh configs/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py 8
## 004 Grayscale Image Deoising (middle size)
./tools/dist_train.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15.py 8
./tools/dist_train.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25.py 8
./tools/dist_train.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50.py 8
## 005 Color Image Deoising (middle size)
./tools/dist_train.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15.py 8
./tools/dist_train.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25.py 8
./tools/dist_train.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50.py 8
## 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding uses 8x8 blocks)
## grayscale
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10.py 8
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20.py 8
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30.py 8
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40.py 8
## color
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10.py 8
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20.py 8
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30.py 8
./tools/dist_train.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
## 001 Classical Image Super-Resolution (middle size)
## (setting1: when model is trained on DIV2K and with training_patch_size=48)
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k-ed2d419e.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k-926950f1.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k-88e4903d.pth
## (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k-69e15fb6.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k-d6982f7b.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k-0502d775.pth
## 002 Lightweight Image Super-Resolution (small size)
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k-131d3f64.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k-309cb239.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k-d6622d03.pth
## 003 Real-World Image Super-Resolution
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-c6425057.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-6f0c425f.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-36960d18.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-a016a72f.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-os-9f1599b5.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-os-25f1722a.pth
## 004 Grayscale Image Deoising (middle size)
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15-6782691b.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25-d0d8d4da.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50-54c9968a.pth
## 005 Color Image Deoising (middle size)
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15-c74a2cee.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25-df2b1c0c.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50-e369874c.pth
## 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding usesx8 blocks)
## grayscale
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10-da93c8e9.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20-d47367b1.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30-52c083cf.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40-803e8d9b.pth
## color
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10-09aafadc.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20-b8a42b5e.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30-e9fe6859.pth
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40-5b77a6e6.pth
## single-gpu test
## 001 Classical Image Super-Resolution (middle size)
## (setting1: when model is trained on DIV2K and with training_patch_size=48)
python tools/test.py configs/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k-ed2d419e.pth
python tools/test.py configs/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k-926950f1.pth
python tools/test.py configs/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k-88e4903d.pth
## (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
python tools/test.py configs/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k-69e15fb6.pth
python tools/test.py configs/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k-d6982f7b.pth
python tools/test.py configs/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k-0502d775.pth
## 002 Lightweight Image Super-Resolution (small size)
python tools/test.py configs/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k-131d3f64.pth
python tools/test.py configs/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k-309cb239.pth
python tools/test.py configs/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k-d6622d03.pth
## 003 Real-World Image Super-Resolution
python tools/test.py configs/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-c6425057.pth
python tools/test.py configs/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-6f0c425f.pth
python tools/test.py configs/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-36960d18.pth
python tools/test.py configs/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-a016a72f.pth
python tools/test.py configs/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-os-9f1599b5.pth
python tools/test.py configs/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-os-25f1722a.pth
## 004 Grayscale Image Deoising (middle size)
python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15-6782691b.pth
python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25-d0d8d4da.pth
python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50-54c9968a.pth
## 005 Color Image Deoising (middle size)
python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15-c74a2cee.pth
python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25-df2b1c0c.pth
python tools/test.py configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50-e369874c.pth
## 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding usesx8 blocks)
## grayscale
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10-da93c8e9.pth
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20-d47367b1.pth
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30-52c083cf.pth
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40-803e8d9b.pth
## color
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10-09aafadc.pth
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20-b8a42b5e.pth
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30-e9fe6859.pth
python tools/test.py configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40-5b77a6e6.pth
## multi-gpu test
## 001 Classical Image Super-Resolution (middle size)
## (setting1: when model is trained on DIV2K and with training_patch_size=48)
./tools/dist_test.sh configs/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s48w8d6e180_8xb4-lr2e-4-500k_div2k-ed2d419e.pth
./tools/dist_test.sh configs/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s48w8d6e180_8xb4-lr2e-4-500k_div2k-926950f1.pth
./tools/dist_test.sh configs/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s48w8d6e180_8xb4-lr2e-4-500k_div2k-88e4903d.pth
## (setting2: when model is trained on DIV2K+Flickr2K and with training_patch_size=64)
./tools/dist_test.sh configs/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s64w8d6e180_8xb4-lr2e-4-500k_df2k-69e15fb6.pth
./tools/dist_test.sh configs/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s64w8d6e180_8xb4-lr2e-4-500k_df2k-d6982f7b.pth
./tools/dist_test.sh configs/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s64w8d6e180_8xb4-lr2e-4-500k_df2k-0502d775.pth
## 002 Lightweight Image Super-Resolution (small size)
./tools/dist_test.sh configs/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x2s64w8d4e60_8xb4-lr2e-4-500k_div2k-131d3f64.pth
./tools/dist_test.sh configs/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x3s64w8d4e60_8xb4-lr2e-4-500k_div2k-309cb239.pth
./tools/dist_test.sh configs/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k.py https://download.openmmlab.com/mmediting/swinir/swinir_x4s64w8d4e60_8xb4-lr2e-4-500k_div2k-d6622d03.pth
## 003 Real-World Image Super-Resolution
./tools/dist_test.sh configs/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-c6425057.pth
./tools/dist_test.sh configs/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x2s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-6f0c425f.pth
./tools/dist_test.sh configs/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-36960d18.pth
./tools/dist_test.sh configs/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x4s64w8d6e180_8xb4-lr1e-4-600k_df2k-os-a016a72f.pth
./tools/dist_test.sh configs/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_gan-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-os-9f1599b5.pth
./tools/dist_test.sh configs/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-ost.py https://download.openmmlab.com/mmediting/swinir/swinir_psnr-x4s64w8d9e240_8xb4-lr1e-4-600k_df2k-os-25f1722a.pth
## 004 Grayscale Image Deoising (middle size)
./tools/dist_test.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN15-6782691b.pth
./tools/dist_test.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN25-d0d8d4da.pth
./tools/dist_test.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-grayDN50-54c9968a.pth
## 005 Color Image Deoising (middle size)
./tools/dist_test.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN15-c74a2cee.pth
./tools/dist_test.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN25-df2b1c0c.pth
./tools/dist_test.sh configs/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50.py https://download.openmmlab.com/mmediting/swinir/swinir_s128w8d6e180_8xb1-lr2e-4-1600k_dfwb-colorDN50-e369874c.pth
## 006 JPEG Compression Artifact Reduction (middle size, using window_size=7 because JPEG encoding uses 8x8 blocks)
## grayscale
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR10-da93c8e9.pth
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR20-d47367b1.pth
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR30-52c083cf.pth
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-grayCAR40-803e8d9b.pth
## color
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR10-09aafadc.pth
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR20-b8a42b5e.pth
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30.py https://download.openmmlab.com/mmediting/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR30-e9fe6859.pth
./tools/dist_test.sh configs/swinir/swinir_s126w7d6e180_8xb1-lr2e-4-1600k_dfwb-colorCAR40.py https://download.openmmlab.com/mmediting/swinir/
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{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},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={1833--1844},
year={2021}
}
LIIF (CVPR’2021)¶
Task: Image Super-Resolution
Abstract¶
How to represent an image? While the visual world is presented in a continuous manner, machines store and see the images in a discrete way with 2D arrays of pixels. In this paper, we seek to learn a continuous representation for images. Inspired by the recent progress in 3D reconstruction with implicit neural representation, we propose Local Implicit Image Function (LIIF), which takes an image coordinate and the 2D deep features around the coordinate as inputs, predicts the RGB value at a given coordinate as an output. Since the coordinates are continuous, LIIF can be presented in arbitrary resolution. To generate the continuous representation for images, we train an encoder with LIIF representation via a self-supervised task with super-resolution. The learned continuous representation can be presented in arbitrary resolution even extrapolate to x30 higher resolution, where the training tasks are not provided. We further show that LIIF representation builds a bridge between discrete and continuous representation in 2D, it naturally supports the learning tasks with size-varied image ground-truths and significantly outperforms the method with resizing the ground-truths.
Results and models¶
Note:
△ refers to ditto.
Evaluated on RGB channels,
scalepixels in each border are cropped before evaluation.
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/liif/liif-edsr-norm_c64b16_1xb16-1000k_div2k.py
## single-gpu train
python tools/train.py configs/liif/liif-edsr-norm_c64b16_1xb16-1000k_div2k.py
## multi-gpu train
./tools/dist_train.sh configs/liif/liif-edsr-norm_c64b16_1xb16-1000k_div2k.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/liif/liif-edsr-norm_c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/liif/liif_edsr_norm_c64b16_g1_1000k_div2k_20210715-ab7ce3fc.pth
## single-gpu test
python tools/test.py configs/liif/liif-edsr-norm_c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/liif/liif_edsr_norm_c64b16_g1_1000k_div2k_20210715-ab7ce3fc.pth
## multi-gpu test
./tools/dist_test.sh configs/liif/liif-edsr-norm_c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/liif/liif_edsr_norm_c64b16_g1_1000k_div2k_20210715-ab7ce3fc.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{chen2021learning,
title={Learning continuous image representation with local implicit image function},
author={Chen, Yinbo and Liu, Sifei and Wang, Xiaolong},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={8628--8638},
year={2021}
}
GLEAN (CVPR’2021)¶
Task: Image Super-Resolution
Abstract¶
We show that pre-trained Generative Adversarial Networks (GANs), e.g., StyleGAN, can be used as a latent bank to improve the restoration quality of large-factor image super-resolution (SR). While most existing SR approaches attempt to generate realistic textures through learning with adversarial loss, our method, Generative LatEnt bANk (GLEAN), goes beyond existing practices by directly leveraging rich and diverse priors encapsulated in a pre-trained GAN. But unlike prevalent GAN inversion methods that require expensive image-specific optimization at runtime, our approach only needs a single forward pass to generate the upscaled image. GLEAN can be easily incorporated in a simple encoder-bank-decoder architecture with multi-resolution skip connections. Switching the bank allows the method to deal with images from diverse categories, e.g., cat, building, human face, and car. Images upscaled by GLEAN show clear improvements in terms of fidelity and texture faithfulness in comparison to existing methods.
Results and models¶
For the meta info used in training and test, please refer to here. The results are evaluated on RGB channels.
| Model | Dataset | PSNR | Training Resources | Download |
|---|---|---|---|---|
| glean_cat_8x | LSUN-CAT | 23.98 | 2 (Tesla V100-PCIE-32GB) | model | log |
| glean_ffhq_16x | FFHQ | 26.91 | 2 (Tesla V100-PCIE-32GB) | model | log |
| glean_cat_16x | LSUN-CAT | 20.88 | 2 (Tesla V100-PCIE-32GB) | model | log |
| glean_in128out1024_4x2_300k_ffhq_celebahq | FFHQ | 27.94 | 4 (Tesla V100-SXM3-32GB) | model | log |
| glean_fp16_cat_8x | LSUN-CAT | - | - | - |
| glean_fp16_ffhq_16x | FFHQ | - | - | - |
| glean_fp16_in128out1024_4x2_300k_ffhq_celebahq | FFHQ | - | - | - |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/glean/glean_x8_2xb8_cat.py
## single-gpu train
python tools/train.py configs/glean/glean_x8_2xb8_cat.py
## multi-gpu train
./tools/dist_train.sh configs/glean/glean_x8_2xb8_cat.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/glean/glean_x8_2xb8_cat.py https://download.openmmlab.com/mmediting/restorers/glean/glean_cat_8x_20210614-d3ac8683.pth
## single-gpu test
python tools/test.py configs/glean/glean_x8_2xb8_cat.py https://download.openmmlab.com/mmediting/restorers/glean/glean_cat_8x_20210614-d3ac8683.pth
## multi-gpu test
./tools/dist_test.sh configs/glean/glean_x8_2xb8_cat.py https://download.openmmlab.com/mmediting/restorers/glean/glean_cat_8x_20210614-d3ac8683.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@InProceedings{chan2021glean,
author = {Chan, Kelvin CK and Wang, Xintao and Xu, Xiangyu and Gu, Jinwei and Loy, Chen Change},
title = {GLEAN: Generative Latent Bank for Large-Factor Image Super-Resolution},
booktitle = {Proceedings of the IEEE conference on computer vision and pattern recognition},
year = {2021}
}
Real-ESRGAN (ICCVW’2021)¶
Task: Image Super-Resolution
Abstract¶
Though many attempts have been made in blind super-resolution to restore low-resolution images with unknown and complex degradations, they are still far from addressing general real-world degraded images. In this work, we extend the powerful ESRGAN to a practical restoration application (namely, Real-ESRGAN), which is trained with pure synthetic data. Specifically, a high-order degradation modeling process is introduced to better simulate complex real-world degradations. We also consider the common ringing and overshoot artifacts in the synthesis process. In addition, we employ a U-Net discriminator with spectral normalization to increase discriminator capability and stabilize the training dynamics. Extensive comparisons have shown its superior visual performance than prior works on various real datasets. We also provide efficient implementations to synthesize training pairs on the fly.
Results and models¶
Evaluated on Set5 dataset with RGB channels. The metrics are PSNR and SSIM.
| Model | Dataset | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|
| realesrnet_c64b23g32_12x4_lr2e-4_1000k_df2k_ost | df2k_ost | 28.0297 | 0.8236 | 4 (Tesla V100-SXM2-32GB) | model/log |
| realesrgan_c64b23g32_12x4_lr1e-4_400k_df2k_ost | df2k_ost | 26.2204 | 0.7655 | 4 (Tesla V100-SXM2-32GB) | model /log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/real_esrgan/realesrgan_c64b23g32_4xb12-lr1e-4-400k_df2k-ost.py
## single-gpu train
python tools/train.py configs/real_esrgan/realesrgan_c64b23g32_4xb12-lr1e-4-400k_df2k-ost.py
## multi-gpu train
./tools/dist_train.sh configs/real_esrgan/realesrgan_c64b23g32_4xb12-lr1e-4-400k_df2k-ost.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/real_esrgan/realesrgan_c64b23g32_4xb12-lr1e-4-400k_df2k-ost.py https://download.openmmlab.com/mmediting/restorers/real_esrgan/realesrgan_c64b23g32_12x4_lr1e-4_400k_df2k_ost_20211010-34798885.pth
## single-gpu test
python tools/test.py configs/real_esrgan/realesrgan_c64b23g32_4xb12-lr1e-4-400k_df2k-ost.py https://download.openmmlab.com/mmediting/restorers/real_esrgan/realesrgan_c64b23g32_12x4_lr1e-4_400k_df2k_ost_20211010-34798885.pth
## multi-gpu test
./tools/dist_test.sh configs/real_esrgan/realesrgan_c64b23g32_4xb12-lr1e-4-400k_df2k-ost.py https://download.openmmlab.com/mmediting/restorers/real_esrgan/realesrgan_c64b23g32_12x4_lr1e-4_400k_df2k_ost_20211010-34798885.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{wang2021real,
title={Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic data},
author={Wang, Xintao and Xie, Liangbin and Dong, Chao and Shan, Ying},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)},
pages={1905--1914},
year={2021}
}
TTSR (CVPR’2020)¶
Task: Image Super-Resolution
Abstract¶
We study on image super-resolution (SR), which aims to recover realistic textures from a low-resolution (LR) image. Recent progress has been made by taking high-resolution images as references (Ref), so that relevant textures can be transferred to LR images. However, existing SR approaches neglect to use attention mechanisms to transfer high-resolution (HR) textures from Ref images, which limits these approaches in challenging cases. In this paper, we propose a novel Texture Transformer Network for Image Super-Resolution (TTSR), in which the LR and Ref images are formulated as queries and keys in a transformer, respectively. TTSR consists of four closely-related modules optimized for image generation tasks, including a learnable texture extractor by DNN, a relevance embedding module, a hard-attention module for texture transfer, and a soft-attention module for texture synthesis. Such a design encourages joint feature learning across LR and Ref images, in which deep feature correspondences can be discovered by attention, and thus accurate texture features can be transferred. The proposed texture transformer can be further stacked in a cross-scale way, which enables texture recovery from different levels (e.g., from 1x to 4x magnification). Extensive experiments show that TTSR achieves significant improvements over state-of-the-art approaches on both quantitative and qualitative evaluations.
Results and models¶
Evaluated on CUFED dataset (RGB channels), scale pixels in each border are cropped before evaluation.
The metrics are PSNR and SSIM .
| Model | Dataset | scale | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|---|
| ttsr-rec_x4_c64b16_g1_200k_CUFED | CUFED | x4 | 25.2433 | 0.7491 | 1 (TITAN Xp) | model | log |
| ttsr-gan_x4_c64b16_g1_500k_CUFED | CUFED | x4 | 24.6075 | 0.7234 | 1 (TITAN Xp) | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/ttsr/ttsr-gan_x4c64b16_1xb9-500k_CUFED.py
## single-gpu train
python tools/train.py configs/ttsr/ttsr-gan_x4c64b16_1xb9-500k_CUFED.py
## multi-gpu train
./tools/dist_train.sh configs/ttsr/ttsr-gan_x4c64b16_1xb9-500k_CUFED.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/ttsr/ttsr-gan_x4c64b16_1xb9-500k_CUFED.py https://download.openmmlab.com/mmediting/restorers/ttsr/ttsr-gan_x4_c64b16_g1_500k_CUFED_20210626-2ab28ca0.pth
## single-gpu test
python tools/test.py configs/ttsr/ttsr-gan_x4c64b16_1xb9-500k_CUFED.py https://download.openmmlab.com/mmediting/restorers/ttsr/ttsr-gan_x4_c64b16_g1_500k_CUFED_20210626-2ab28ca0.pth
## multi-gpu test
./tools/dist_test.sh configs/ttsr/ttsr-gan_x4c64b16_1xb9-500k_CUFED.py https://download.openmmlab.com/mmediting/restorers/ttsr/ttsr-gan_x4_c64b16_g1_500k_CUFED_20210626-2ab28ca0.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{yang2020learning,
title={Learning texture transformer network for image super-resolution},
author={Yang, Fuzhi and Yang, Huan and Fu, Jianlong and Lu, Hongtao and Guo, Baining},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5791--5800},
year={2020}
}
DIC (CVPR’2020)¶
Task: Image Super-Resolution
Abstract¶
Recent works based on deep learning and facial priors have succeeded in super-resolving severely degraded facial images. However, the prior knowledge is not fully exploited in existing methods, since facial priors such as landmark and component maps are always estimated by low-resolution or coarsely super-resolved images, which may be inaccurate and thus affect the recovery performance. In this paper, we propose a deep face super-resolution (FSR) method with iterative collaboration between two recurrent networks which focus on facial image recovery and landmark estimation respectively. In each recurrent step, the recovery branch utilizes the prior knowledge of landmarks to yield higher-quality images which facilitate more accurate landmark estimation in turn. Therefore, the iterative information interaction between two processes boosts the performance of each other progressively. Moreover, a new attentive fusion module is designed to strengthen the guidance of landmark maps, where facial components are generated individually and aggregated attentively for better restoration. Quantitative and qualitative experimental results show the proposed method significantly outperforms state-of-the-art FSR methods in recovering high-quality face images.
Results and models¶
Evaluated on RGB channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
In the log data of dic_gan_x8c48b6_g4_150k_CelebAHQ, DICGAN is verified on the first 9 pictures of the test set of CelebA-HQ, so PSNR and SSIM shown in the follow table is different from the log data.
Training Resources: Training Resourcesrmation during training.
| Model | Dataset | scale | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|---|
| dic_x8c48b6_g4_150k_CelebAHQ | CelebAHQ | x8 | 25.2319 | 0.7422 | 4 (Tesla PG503-216) | model | log |
| dic_gan_x8c48b6_g4_500k_CelebAHQ | CelebAHQ | x8 | 23.6241 | 0.6721 | 4 (Tesla PG503-216) | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/dic/dic_gan-x8c48b6_4xb2-500k_celeba-hq.py
## single-gpu train
python tools/train.py configs/dic/dic_gan-x8c48b6_4xb2-500k_celeba-hq.py
## multi-gpu train
./tools/dist_train.sh configs/dic/dic_gan-x8c48b6_4xb2-500k_celeba-hq.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/dic/dic_gan-x8c48b6_4xb2-500k_celeba-hq.py https://download.openmmlab.com/mmediting/restorers/dic/dic_gan_x8c48b6_g4_500k_CelebAHQ_20210625-3b89a358.pth
## single-gpu test
python tools/test.py configs/dic/dic_gan-x8c48b6_4xb2-500k_celeba-hq.py https://download.openmmlab.com/mmediting/restorers/dic/dic_gan_x8c48b6_g4_500k_CelebAHQ_20210625-3b89a358.pth
## multi-gpu test
./tools/dist_test.sh configs/dic/dic_gan-x8c48b6_4xb2-500k_celeba-hq.py https://download.openmmlab.com/mmediting/restorers/dic/dic_gan_x8c48b6_g4_500k_CelebAHQ_20210625-3b89a358.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{ma2020deep,
title={Deep face super-resolution with iterative collaboration between attentive recovery and landmark estimation},
author={Ma, Cheng and Jiang, Zhenyu and Rao, Yongming and Lu, Jiwen and Zhou, Jie},
booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
pages={5569--5578},
year={2020}
}
RDN (CVPR’2018)¶
Task: Image Super-Resolution
Abstract¶
A very deep convolutional neural network (CNN) has recently achieved great success for image super-resolution (SR) and offered hierarchical features as well. However, most deep CNN based SR models do not make full use of the hierarchical features from the original low-resolution (LR) images, thereby achieving relatively-low performance. In this paper, we propose a novel residual dense network (RDN) to address this problem in image SR. We fully exploit the hierarchical features from all the convolutional layers. Specifically, we propose residual dense block (RDB) to extract abundant local features via dense connected convolutional layers. RDB further allows direct connections from the state of preceding RDB to all the layers of current RDB, leading to a contiguous memory (CM) mechanism. Local feature fusion in RDB is then used to adaptively learn more effective features from preceding and current local features and stabilizes the training of wider network. After fully obtaining dense local features, we use global feature fusion to jointly and adaptively learn global hierarchical features in a holistic way. Extensive experiments on benchmark datasets with different degradation models show that our RDN achieves favorable performance against state-of-the-art methods.
Results and models¶
Evaluated on RGB channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR and SSIM .
| Model | Dataset | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|
| rdn_x4c64b16_g1_1000k_div2k | Set5 | 30.4922 | 0.8548 | 1 (TITAN Xp) | model | log |
| rdn_x4c64b16_g1_1000k_div2k | Set14 | 26.9570 | 0.7423 | 1 (TITAN Xp) | model | log |
| rdn_x4c64b16_g1_1000k_div2k | DIV2K | 29.1925 | 0.8233 | 1 (TITAN Xp) | model | log |
| rdn_x3c64b16_g1_1000k_div2k | Set5 | 32.6051 | 0.8943 | 1 (TITAN Xp) | model | log |
| rdn_x3c64b16_g1_1000k_div2k | Set14 | 28.6338 | 0.8077 | 1 (TITAN Xp) | model | log |
| rdn_x3c64b16_g1_1000k_div2k | DIV2K | 31.2153 | 0.8763 | 1 (TITAN Xp) | model | log |
| rdn_x2c64b16_g1_1000k_div2k | Set5 | 35.9883 | 0.9385 | 1 (TITAN Xp) | model | log |
| rdn_x2c64b16_g1_1000k_div2k | Set14 | 31.8366 | 0.8920 | 1 (TITAN Xp) | model | log |
| rdn_x2c64b16_g1_1000k_div2k | DIV2K | 34.9392 | 0.9380 | 1 (TITAN Xp) | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/rdn/rdn_x4c64b16_1xb16-1000k_div2k.py
## single-gpu train
python tools/train.py configs/rdn/rdn_x4c64b16_1xb16-1000k_div2k.py
## multi-gpu train
./tools/dist_train.sh configs/rdn/rdn_x4c64b16_1xb16-1000k_div2k.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/rdn/rdn_x4c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/rdn/rdn_x4c64b16_g1_1000k_div2k_20210419-3577d44f.pth
## single-gpu test
python tools/test.py configs/rdn/rdn_x4c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/rdn/rdn_x4c64b16_g1_1000k_div2k_20210419-3577d44f.pth
## multi-gpu test
./tools/dist_test.sh configs/rdn/rdn_x4c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/rdn/rdn_x4c64b16_g1_1000k_div2k_20210419-3577d44f.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{zhang2018residual,
title={Residual dense network for image super-resolution},
author={Zhang, Yulun and Tian, Yapeng and Kong, Yu and Zhong, Bineng and Fu, Yun},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={2472--2481},
year={2018}
}
ESRGAN (ECCVW’2018)¶
Task: Image Super-Resolution
Abstract¶
The Super-Resolution Generative Adversarial Network (SRGAN) is a seminal work that is capable of generating realistic textures during single image super-resolution. However, the hallucinated details are often accompanied with unpleasant artifacts. To further enhance the visual quality, we thoroughly study three key components of SRGAN - network architecture, adversarial loss and perceptual loss, and improve each of them to derive an Enhanced SRGAN (ESRGAN). In particular, we introduce the Residual-in-Residual Dense Block (RRDB) without batch normalization as the basic network building unit. Moreover, we borrow the idea from relativistic GAN to let the discriminator predict relative realness instead of the absolute value. Finally, we improve the perceptual loss by using the features before activation, which could provide stronger supervision for brightness consistency and texture recovery. Benefiting from these improvements, the proposed ESRGAN achieves consistently better visual quality with more realistic and natural textures than SRGAN and won the first place in the PIRM2018-SR Challenge.
Results and models¶
Evaluated on RGB channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
| Model | Dataset | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|
| esrgan_psnr_x4c64b23g32_1x16_1000k_div2k | Set5 | 30.6428 | 0.8559 | 1 | model |
| esrgan_psnr_x4c64b23g32_1x16_1000k_div2k | Set14 | 27.0543 | 0.7447 | 1 | model |
| esrgan_psnr_x4c64b23g32_1x16_1000k_div2k | DIV2K | 29.3354 | 0.8263 | 1 | model |
| esrgan_x4c64b23g32_1x16_400k_div2k | Set5 | 28.2700 | 0.7778 | 1 | model |
| esrgan_x4c64b23g32_1x16_400k_div2k | Set14 | 24.6328 | 0.6491 | 1 | model |
| esrgan_x4c64b23g32_1x16_400k_div2k | DIV2K | 26.6531 | 0.7340 | 1 | model |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/esrgan/esrgan_x4c64b23g32_1xb16-400k_div2k.py
## single-gpu train
python tools/train.py configs/esrgan/esrgan_x4c64b23g32_1xb16-400k_div2k.py
## multi-gpu train
./tools/dist_train.sh configs/esrgan/esrgan_x4c64b23g32_1xb16-400k_div2k.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/esrgan/esrgan_x4c64b23g32_1xb16-400k_div2k.py https://download.openmmlab.com/mmediting/restorers/esrgan/esrgan_x4c64b23g32_1x16_400k_div2k_20200508-f8ccaf3b.pth
## single-gpu test
python tools/test.py configs/esrgan/esrgan_x4c64b23g32_1xb16-400k_div2k.py https://download.openmmlab.com/mmediting/restorers/esrgan/esrgan_x4c64b23g32_1x16_400k_div2k_20200508-f8ccaf3b.pth
## multi-gpu test
./tools/dist_test.sh configs/esrgan/esrgan_x4c64b23g32_1xb16-400k_div2k.py https://download.openmmlab.com/mmediting/restorers/esrgan/esrgan_x4c64b23g32_1x16_400k_div2k_20200508-f8ccaf3b.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{wang2018esrgan,
title={Esrgan: Enhanced super-resolution generative adversarial networks},
author={Wang, Xintao and Yu, Ke and Wu, Shixiang and Gu, Jinjin and Liu, Yihao and Dong, Chao and Qiao, Yu and Change Loy, Chen},
booktitle={Proceedings of the European Conference on Computer Vision Workshops(ECCVW)},
pages={0--0},
year={2018}
}
EDSR (CVPR’2017)¶
Task: Image Super-Resolution
Abstract¶
Recent research on super-resolution has progressed with the development of deep convolutional neural networks (DCNN). In particular, residual learning techniques exhibit improved performance. In this paper, we develop an enhanced deep super-resolution network (EDSR) with performance exceeding those of current state-of-the-art SR methods. The significant performance improvement of our model is due to optimization by removing unnecessary modules in conventional residual networks. The performance is further improved by expanding the model size while we stabilize the training procedure. We also propose a new multi-scale deep super-resolution system (MDSR) and training method, which can reconstruct high-resolution images of different upscaling factors in a single model. The proposed methods show superior performance over the state-of-the-art methods on benchmark datasets and prove its excellence by winning the NTIRE2017 Super-Resolution Challenge.
Results and models¶
Evaluated on RGB channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
| Model | Dataset | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|
| edsr_x2c64b16_1x16_300k_div2k | Set5 | 35.7592 | 0.9372 | 1 | model | log |
| edsr_x2c64b16_1x16_300k_div2k | Set14 | 31.4290 | 0.8874 | 1 | model | log |
| edsr_x2c64b16_1x16_300k_div2k | DIV2K | 34.5896 | 0.9352 | 1 | model | log |
| edsr_x3c64b16_1x16_300k_div2k | Set5 | 32.3301 | 0.8912 | 1 | model | log |
| edsr_x3c64b16_1x16_300k_div2k | Set14 | 28.4125 | 0.8022 | 1 | model | log |
| edsr_x3c64b16_1x16_300k_div2k | DIV2K | 30.9154 | 0.8711 | 1 | model | log |
| edsr_x4c64b16_1x16_300k_div2k | Set5 | 30.2223 | 0.8500 | 1 | model | log |
| edsr_x4c64b16_1x16_300k_div2k | Set14 | 26.7870 | 0.7366 | 1 | model | log |
| edsr_x4c64b16_1x16_300k_div2k | DIV2K | 28.9675 | 0.8172 | 1 | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/edsr/edsr_x4c64b16_1xb16-300k_div2k.py
## single-gpu train
python tools/train.py configs/edsr/edsr_x4c64b16_1xb16-300k_div2k.py
## multi-gpu train
./tools/dist_train.sh configs/edsr/edsr_x4c64b16_1xb16-300k_div2k.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/edsr/edsr_x4c64b16_1xb16-300k_div2k.py https://download.openmmlab.com/mmediting/restorers/edsr/edsr_x4c64b16_1x16_300k_div2k_20200608-3c2af8a3.pth
## single-gpu test
python tools/test.py configs/edsr/edsr_x4c64b16_1xb16-300k_div2k.py https://download.openmmlab.com/mmediting/restorers/edsr/edsr_x4c64b16_1x16_300k_div2k_20200608-3c2af8a3.pth
## multi-gpu test
./tools/dist_test.sh configs/edsr/edsr_x4c64b16_1xb16-300k_div2k.py https://download.openmmlab.com/mmediting/restorers/edsr/edsr_x4c64b16_1x16_300k_div2k_20200608-3c2af8a3.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{lim2017enhanced,
title={Enhanced deep residual networks for single image super-resolution},
author={Lim, Bee and Son, Sanghyun and Kim, Heewon and Nah, Seungjun and Mu Lee, Kyoung},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition workshops},
pages={136--144},
year={2017}
}
SRGAN (CVPR’2016)¶
Task: Image Super-Resolution
Abstract¶
Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.
Results and models¶
Evaluated on RGB channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
| Model |Dataset| PSNR | SSIM | Training Resources | Download | | :——-: | :——–: | :——–: | :——-: | :——–: | :——–: | :—————-: | | msrresnet_x4c64b16_1x16_300k_div2k | Set5|30.2252 | 0.8491 | 1 | model | log | | msrresnet_x4c64b16_1x16_300k_div2k | Set14|26.7762 | 0.7369 | 1 | model | log | | msrresnet_x4c64b16_1x16_300k_div2k | DIV2K|28.9748 | 0.8178 | 1 | model | log | | srgan_x4c64b16_1x16_1000k_div2k |Set5|27.9499 | 0.7846 |1 | model | log | | srgan_x4c64b16_1x16_1000k_div2k |Set14|24.7383 |0.6491 | 1 | model | log | | srgan_x4c64b16_1x16_1000k_div2k |DIV2K|26.5697 |0.7365 | 1 | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/srgan_resnet/srgan_x4c64b16_1xb16-1000k_div2k.py
## single-gpu train
python tools/train.py configs/srgan_resnet/srgan_x4c64b16_1xb16-1000k_div2k.py
## multi-gpu train
./tools/dist_train.sh configs/srgan_resnet/srgan_x4c64b16_1xb16-1000k_div2k.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/srgan_resnet/srgan_x4c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/srresnet_srgan/srgan_x4c64b16_1x16_1000k_div2k_20200606-a1f0810e.pth
## single-gpu test
python tools/test.py configs/srgan_resnet/srgan_x4c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/srresnet_srgan/srgan_x4c64b16_1x16_1000k_div2k_20200606-a1f0810e.pth
## multi-gpu test
./tools/dist_test.sh configs/srgan_resnet/srgan_x4c64b16_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/srresnet_srgan/srgan_x4c64b16_1x16_1000k_div2k_20200606-a1f0810e.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@inproceedings{ledig2016photo,
title={Photo-realistic single image super-resolution using a generative adversarial network},
author={Ledig, Christian and Theis, Lucas and Husz{\'a}r, Ferenc and Caballero, Jose and Cunningham, Andrew and Acosta, Alejandro and Aitken, Andrew and Tejani, Alykhan and Totz, Johannes and Wang, Zehan},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition workshops},
year={2016}
}
SRCNN (TPAMI’2015)¶
Task: Image Super-Resolution
Abstract¶
We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low/high-resolution images. The mapping is represented as a deep convolutional neural network (CNN) that takes the low-resolution image as the input and outputs the high-resolution one. We further show that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network. But unlike traditional methods that handle each component separately, our method jointly optimizes all layers. Our deep CNN has a lightweight structure, yet demonstrates state-of-the-art restoration quality, and achieves fast speed for practical on-line usage. We explore different network structures and parameter settings to achieve trade-offs between performance and speed. Moreover, we extend our network to cope with three color channels simultaneously, and show better overall reconstruction quality.
Results and models¶
Evaluated on RGB channels, scale pixels in each border are cropped before evaluation.
The metrics are PSNR / SSIM .
| Model | Dataset | PSNR | SSIM | Training Resources | Download |
|---|---|---|---|---|---|
| srcnn_x4k915_1x16_1000k_div2k | Set5 | 28.4316 | 0.8099 | 1 | model | log |
| srcnn_x4k915_1x16_1000k_div2k | Set14 | 25.6486 | 0.7014 | 1 | model | log |
| srcnn_x4k915_1x16_1000k_div2k | DIV2K | 27.7460 | 0.7854 | 1 | model | log |
Quick Start¶
Train
Train Instructions
You can use the following commands to train a model with cpu or single/multiple GPUs.
## cpu train
CUDA_VISIBLE_DEVICES=-1 python tools/train.py configs/srcnn/srcnn_x4k915_1xb16-1000k_div2k.py
## single-gpu train
python tools/train.py configs/srcnn/srcnn_x4k915_1xb16-1000k_div2k.py
## multi-gpu train
./tools/dist_train.sh configs/srcnn/srcnn_x4k915_1xb16-1000k_div2k.py 8
For more details, you can refer to Train a model part in train_test.md.
Test
Test Instructions
You can use the following commands to test a model with cpu or single/multiple GPUs.
## cpu test
CUDA_VISIBLE_DEVICES=-1 python tools/test.py configs/srcnn/srcnn_x4k915_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/srcnn/srcnn_x4k915_1x16_1000k_div2k_20200608-4186f232.pth
## single-gpu test
python tools/test.py configs/srcnn/srcnn_x4k915_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/srcnn/srcnn_x4k915_1x16_1000k_div2k_20200608-4186f232.pth
## multi-gpu test
./tools/dist_test.sh configs/srcnn/srcnn_x4k915_1xb16-1000k_div2k.py https://download.openmmlab.com/mmediting/restorers/srcnn/srcnn_x4k915_1x16_1000k_div2k_20200608-4186f232.pth 8
For more details, you can refer to Test a pre-trained model part in train_test.md.
Citation¶
@article{dong2015image,
title={Image super-resolution using deep convolutional networks},
author={Dong, Chao and Loy, Chen Change and He, Kaiming and Tang, Xiaoou},
journal={IEEE transactions on pattern analysis and machine intelligence},
volume={38},
number={2},
pages={295--307},
year={2015},
publisher={IEEE}
}