File size: 7,443 Bytes
a16c423 cd43d33 a16c423 eea1706 a16c423 eea1706 a16c423 ca3844d cd43d33 ca3844d cd43d33 a16c423 eea1706 a16c423 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 |
---
license: apache-2.0
tags:
- super-image
- image-super-resolution
datasets:
- eugenesiow/Div2k
metrics:
- pnsr
- ssim
---
# Multi-scale Residual Network for Image Super-Resolution (MSRN)
MSRN model pre-trained on DIV2K (800 images training, augmented to 4000 images, 100 images validation) for 2x, 3x and 4x image super resolution. It was introduced in the paper [Multi-scale Residual Network for Image Super-Resolution](https://openaccess.thecvf.com/content_ECCV_2018/html/Juncheng_Li_Multi-scale_Residual_Network_ECCV_2018_paper.html) by Li et al. (2018) and first released in [this repository](https://github.com/MIVRC/MSRN-PyTorch).
The goal of image super resolution is to restore a high resolution (HR) image from a single low resolution (LR) image. The image below shows the ground truth (HR), the bicubic upscaling x2 and model upscaling x2.
## Model description
The MSRN model proposes a feature extraction structure called the multi-scale residual block. This module can "adaptively detect image features at different scales" and "exploit the potential features of the image".
This model also applies the balanced attention (BAM) method invented by [Wang et al. (2021)](https://arxiv.org/abs/2104.07566) to further improve the results.
## Intended uses & limitations
You can use the pre-trained models for upscaling your images 2x, 3x and 4x. You can also use the trainer to train a model on your own dataset.
### How to use
The model can be used with the [super_image](https://github.com/eugenesiow/super-image) library:
```bash
pip install super-image
```
Here is how to use a pre-trained model to upscale your image:
```python
from super_image import MsrnModel, ImageLoader
from PIL import Image
import requests
url = 'https://paperswithcode.com/media/datasets/Set5-0000002728-07a9793f_zA3bDjj.jpg'
image = Image.open(requests.get(url, stream=True).raw)
model = MsrnModel.from_pretrained('eugenesiow/msrn-bam', scale=2) # scale 2, 3 and 4 models available
inputs = ImageLoader.load_image(image)
preds = model(inputs)
ImageLoader.save_image(preds, './scaled_2x.png') # save the output 2x scaled image to `./scaled_2x.png`
ImageLoader.save_compare(inputs, preds, './scaled_2x_compare.png') # save an output comparing the super-image with a bicubic scaling
```
## Training data
The models for 2x, 3x and 4x image super resolution were pretrained on [DIV2K](https://data.vision.ee.ethz.ch/cvl/DIV2K/), a dataset of 800 high-quality (2K resolution) images for training, augmented to 4000 images and uses a dev set of 100 validation images (images numbered 801 to 900).
## Training procedure
### Preprocessing
We follow the pre-processing and training method of [Wang et al.](https://arxiv.org/abs/2104.07566).
Low Resolution (LR) images are created by using bicubic interpolation as the resizing method to reduce the size of the High Resolution (HR) images by x2, x3 and x4 times.
During training, RGB patches with size of 64×64 from the LR input are used together with their corresponding HR patches.
Data augmentation is applied to the training set in the pre-processing stage where five images are created from the four corners and center of the original image.
The following code provides some helper functions to preprocess the data.
```python
from super_image.data import EvalDataset, TrainAugmentDataset, DatasetBuilder
DatasetBuilder.prepare(
base_path='./DIV2K/DIV2K_train_HR',
output_path='./div2k_4x_train.h5',
scale=4,
do_augmentation=True
)
DatasetBuilder.prepare(
base_path='./DIV2K/DIV2K_val_HR',
output_path='./div2k_4x_val.h5',
scale=4,
do_augmentation=False
)
train_dataset = TrainAugmentDataset('./div2k_4x_train.h5', scale=4)
val_dataset = EvalDataset('./div2k_4x_val.h5')
```
### Pretraining
The model was trained on GPU. The training code is provided below:
```python
from super_image import Trainer, TrainingArguments, MsrnModel, MsrnConfig
training_args = TrainingArguments(
output_dir='./results', # output directory
num_train_epochs=1000, # total number of training epochs
)
config = MsrnConfig(
scale=4, # train a model to upscale 4x
bam=True, # apply balanced attention to the network
)
model = MsrnModel(config)
trainer = Trainer(
model=model, # the instantiated model to be trained
args=training_args, # training arguments, defined above
train_dataset=train_dataset, # training dataset
eval_dataset=val_dataset # evaluation dataset
)
trainer.train()
```
## Evaluation results
The evaluation metrics include [PSNR](https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio#Quality_estimation_with_PSNR) and [SSIM](https://en.wikipedia.org/wiki/Structural_similarity#Algorithm).
Evaluation datasets include:
- Set5 - [Bevilacqua et al. (2012)](http://people.rennes.inria.fr/Aline.Roumy/results/SR_BMVC12.html)
- Set14 - [Zeyde et al. (2010)](https://sites.google.com/site/romanzeyde/research-interests)
- BSD100 - [Martin et al. (2001)](https://www.eecs.berkeley.edu/Research/Projects/CS/vision/bsds/)
- Urban100 - [Huang et al. (2015)](https://sites.google.com/site/jbhuang0604/publications/struct_sr)
The results columns below are represented below as `PSNR/SSIM`. They are compared against a Bicubic baseline.
|Dataset |Scale |Bicubic |msrn-bam |
|--- |--- |--- |--- |
|Set5 |2x |33.64/0.9292 |**38.02/0.9608** |
|Set5 |3x |30.39/0.8678 |**35.13/0.9408** |
|Set5 |4x |28.42/0.8101 |**32.26/0.8955** |
|Set14 |2x |30.22/0.8683 |**33.73/0.9186** |
|Set14 |3x |27.53/0.7737 |**31.06/0.8588** |
|Set14 |4x |25.99/0.7023 |**28.78/0.7859** |
|BSD100 |2x |29.55/0.8425 |**33.78/0.9253** |
|BSD100 |3x |27.20/0.7382 |**29.65/0.8196** |
|BSD100 |4x |25.96/0.6672 |**28.51/0.7651** |
|Urban100 |2x |26.66/0.8408 |**32.08/0.9276** |
|Urban100 |3x | |**29.26/0.8736** |
|Urban100 |4x |23.14/0.6573 |**26.10/0.7857** |
## BibTeX entry and citation info
```bibtex
@misc{wang2021bam,
title={BAM: A Lightweight and Efficient Balanced Attention Mechanism for Single Image Super Resolution},
author={Fanyi Wang and Haotian Hu and Cheng Shen},
year={2021},
eprint={2104.07566},
archivePrefix={arXiv},
primaryClass={eess.IV}
}
```
```bibtex
@InProceedings{Li_2018_ECCV,
author = {Li, Juncheng and Fang, Faming and Mei, Kangfu and Zhang, Guixu},
title = {Multi-scale Residual Network for Image Super-Resolution},
booktitle = {The European Conference on Computer Vision (ECCV)},
month = {September},
year = {2018}
}
``` |