Efficient Lightweight Image Denoising with Triple Attention Transformer
Authors: Yubo Zhou, Jin Lin, Fangchen Ye, Yanyun Qu, Yuan Xie
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on several benchmarks demonstrate that the proposed LIDFormer achieves a better trade-off between high performance and low computational complexity on realworld image denoising tasks. |
| Researcher Affiliation | Academia | 1School of Informatics, Xiamen University, Fujian, China 2School of Computer Science and Technology, East China Normal University, Shanghai, China |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements or links indicating that the source code for the methodology is openly available. |
| Open Datasets | Yes | our method and conventional denoising methods adopt the same classic denoising dataset SIDD (Abdelhamed, Lin, and Brown 2018) for model training. Moreover, the trained model is evaluated on two publicly available datasets, SIDD (Abdelhamed, Lin, and Brown 2018) and DND (Plotz and Roth 2017). |
| Dataset Splits | No | The paper mentions using datasets for training and evaluation but does not specify explicit training/validation/test dataset splits (e.g., percentages or sample counts for each split). |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types) used for running its experiments. |
| Software Dependencies | No | The paper mentions optimizers and loss functions, but does not specify software dependencies (e.g., programming languages, libraries, or frameworks) with version numbers. |
| Experiment Setup | Yes | The Adam optimizer with β1 = 0.9, β2 = 0.999 and L1 loss are utilized to train the model. The training process takes 300K iterations with the learning rate being initially set to 3e-4. And the learning rate will gradually decrease to 1e-6 by using cosine annealing technique (Loshchilov and Hutter 2016). For iterative learning, 128 × 128 image patches with RGB channels are used to train a lightweight denoising model. The mini-batch size is set to 16. Besides, the resolution of image patches and the batch size are updated at iteration numbers of 92k, 156K, 204K, 240K, and 276K to (160^2, 8), (192^2, 6), (256^2, 4), (320^2, 2), and (384^2, 1), respectively. Horizontal and vertical flipping are implemented for data augmentation. |