Accurate Image Restoration with Attention Retractable Transformer
Authors: Jiale Zhang, Yulun Zhang, Jinjin Gu, Yongbing Zhang, Linghe Kong, Xin Yuan
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on image super-resolution, denoising, and JPEG compression artifact reduction tasks. Experimental results validate that our proposed ART outperforms state-of-the-art methods on various benchmark datasets both quantitatively and visually. |
| Researcher Affiliation | Academia | Jiale Zhang1, Yulun Zhang2 , Jinjin Gu3,4, Yongbing Zhang5, Linghe Kong1 , Xin Yuan6 1Shanghai Jiao Tong University, 2ETH Z urich, 3Shanghai AI Laboratory, 4The University of Sydney, 5Harbin Institute of Technology (Shenzhen), 6Westlake University |
| Pseudocode | No | The paper describes the proposed method using text and diagrams but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We also provide code and models at https://github.com/gladzhang/ART. |
| Open Datasets | Yes | For image SR, following previous works Zhang et al. (2018b); Haris et al. (2018), we use DIV2K Timofte et al. (2017) and Flickr2K Lim et al. (2017) as training data... For image denoising and JPEG CAR, same as Swin IR Liang et al. (2021), we use training data: DIV2K, Flickr2K, BSD500 Arbelaez et al. (2010), and WED Ma et al. (2016). |
| Dataset Splits | No | The paper uses predefined training datasets (e.g., DIV2K, Flickr2K) and test datasets (e.g., Set5, Urban100) but does not explicitly describe how to perform a train/validation/test split or provide specific percentages/counts for a validation set derived from the training data. |
| Hardware Specification | Yes | Our ART is implemented on Py Torch Paszke et al. (2017) with 4 NVIDIA RTX8000 GPUs. |
| Software Dependencies | No | The paper states 'Our ART is implemented on Py Torch Paszke et al. (2017)' but does not specify the version number of PyTorch or any other software dependencies, making it difficult to reproduce the exact software environment. |
| Experiment Setup | Yes | Firstly, the residual group number, DAB number, and SAB number in each group are set as 6, 3, and 3. Secondly, all the convolutional layers are equipped with 3 3 kernel, 1-length stride, and 1-length padding... we set the channel dimension as 180... Thirdly, the window size in DAB is set as 8 and the interval size in SAB is adjustable according to different tasks... Data augmentation is performed on the training data through horizontal flip and random rotation of 90 , 180 , and 270 . Besides, we crop the original images into 64 64 patches as the basic training inputs for image SR, 128 128 patches for image denoising, and 126 126 patches for JPEG CAR. We resize the training batch to 32 for image SR, and 8 for image denoising and JPEG CAR... We choose ADAM Kingma & Ba (2015) to optimize our ART model with β1 = 0.9, β2 = 0.999, and zero weight decay. The initial learning rate is set as 2 10 4 and is reduced by half as the training iteration reaches a certain number. Taking image SR as an example, we train ART for total 500k iterations and adjust learning rate to half when training iterations reach 250k, 400k, 450k, and 475k... |