DRT: A Lightweight Single Image Deraining Recursive Transformer

By Yuanchu Liang, Saeed Anwar, Yang Liu
College of Engineering and Computer Science, The Australian National University
Data61, CSIRO

---

This repo is the official implementation of DRT: A Lightweight Single Image Deraining Recursive Transformer (arxiv).
The paper has been accepeted by NTIRE2022 workshop of CVPR2022.

Abstract

Over parameterization is a common technique in deep learning to help models learn and generalize sufficiently to the given task; nonetheless, this often leads to enormous network structures and consumes considerable computing resources during training. Recent powerful transformer-based deep learning models on vision tasks usually have heavy parameters and bear training difficulty. However, many dense-prediction low-level computer vision tasks, such as rain streak removing, often need to be executed on devices with limited computing power and memory in practice. Hence, we introduce a recursive local window-based self-attention structure with residual connections and propose deraining a recursive transformer (DRT), which enjoys the superiority of the transformer but requires a small amount of computing resources. In particular, through recursive architecture, our proposed model uses only ~1.3% of the number of parameters of the current best performing model in deraining while exceeding the state-of-the-art methods on the Rain100L benchmark by at least 0.33 dB. Ablation studies also investigate the impact of recursions on derain outcomes. Moreover, since the model contains no deliberate design for deraining, it can also be applied to other image restoration tasks. Our experiment shows that it can achieve competitive results on desnowing. The source code and pretrained model can be found at https://github.com/YC-Liang/DRT.

Running

Installations
Note the script is only tested on Ubuntu 20.04.

• Download Python3, script was ran by Python 3.7.11
• Install cuda >= 10.1 with cudnn >= 7.
• Install PyTorch >= 1.7.1 with torchvision >= 0.8.2
• Install timm >= 0.3.2
• Install torch-summary >= 0.8.2

Data Sets
Training: download the following data sets and unzip into the Data folder

• Rain800: click here.
• Rain100L-Train: click here
• Rain100H-Train: click here
• Snow100K-training: click here

Testing: inside the Data folder, run mkdir Test-HiNet.

• Unzip Rain800, Rain100L and Rain100H into the Test-HiNet folder: click here
• Unzip Snow100K-testset into the Data folder: click here

The Data folder should looks something like this:

|Data
|--Rain800
    |-- training
        |-- ...
    |-- ...
|--Rain100L-Train
    |-- rain
        |-- ...
    |-- norain
        |-- ...
|--Rain100H-Train
    |-- rain
        |-- ...
    |-- norain
        |-- ...
|--Snow100K-training
    |-- all
        |...
|--Test-HiNet
    |-- Rain100H
        |-- ...
    |-- Rain100L
        |-- ...
    |-- Test100
        |-- ...
|--Snow100K-testset
    |-- Snow100K-L
        |-- ...
    |-- Snow100K-M
        |-- ...
    |-- Snow100K-S
        |-- ...

Running Instructions
The core content is written in the Jupyter Notebook DRT.ipynb.
To evaluate, change the evaluate folder directory in the Load Data section and set the last line of the notebook to

• run(train_net = False, loadCkp = True, loadBest = True, new_dataset = False)
  

To train from scratch, change the training directory in the Load Data section and set the last line of the notebook to

• run(train_net = True, loadCkp = False, loadBest = False, new_dataset = False)

Results

Quantitative Results

Rain100L

Rain100H

Test100

Realistic

Citations

@InProceedings{Liang_2022_CVPR,
    author    = {Liang, Yuanchu and Anwar, Saeed and Liu, Yang},
    title     = {DRT: A Lightweight Single Image Deraining Recursive Transformer},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2022},
    pages     = {589-598}
}

@misc{https://doi.org/10.48550/arxiv.2204.11385,
  doi = {10.48550/ARXIV.2204.11385},
  url = {https://arxiv.org/abs/2204.11385},
  author = {Liang, Yuanchu and Anwar, Saeed and Liu, Yang},
  keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {DRT: A Lightweight Single Image Deraining Recursive Transformer},
  publisher = {arXiv},
  year = {2022}, 
  copyright = {arXiv.org perpetual, non-exclusive license}
}