Enhancing RAW-to-sRGB with Decoupled Style Structure in Fourier Domain
Authors: Xuanhua He, Tao Hu, Guoli Wang, Zejin Wang, Run Wang, Qian Zhang, Keyu Yan, Ziyi Chen, Rui Li, Chengjun Xie, Jie Zhang, Man Zhou
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted evaluations on two distinct datasets: the ZRR dataset and the MAI dataset (Ignatov et al. 2021)... We utilize reference evaluation metrics, including PSNR, SSIM, MS-SSIM (Wang, Simoncelli, and Bovik 2003), and LPIPS (Zhang et al. 2018). For more experiments results, please refer to the supplementary material. We conducted multiple ablation experiments on the ZRR dataset to validate our method, and conducted experiments on multiple dimensions such as model structure, loss function, and model parameter quantity. |
| Researcher Affiliation | Collaboration | Xuanhua He1,2*, Tao Hu1,2*, Guoli Wang3, Zejin Wang3, Run Wang3, Qian Zhang3, Keyu Yan1,2, Ziyi Chen4, Rui Li1, Chenjun Xie1, Jie Zhang1 , Man Zhou5 1Hefei Institutes of Physical Science, Chinese Academy of Sciences 2University of Science and Technology of China 3Horizon Robotics 4Tencent Technology 5Nanyang Technological University |
| Pseudocode | No | The paper describes the architecture and components (Phase Enhance Subnet, Amplitude Refine Subnet, Color Adaptation Sub Net) but does not provide structured pseudocode or an algorithm block. |
| Open Source Code | Yes | Code will be available at https://github.com/alexhe101/Fourier ISP. |
| Open Datasets | Yes | We conducted evaluations on two distinct datasets: the ZRR dataset and the MAI dataset (Ignatov et al. 2021). The ZRR dataset involves mapping RAW images from the Huawei P20 camera to RGB images from a Canon camera. Meanwhile, the MAI dataset focuses on mapping Sony IMX586 Quad Bayer RAW images to Fuji camera RGB images. |
| Dataset Splits | No | The paper states "for both training and testing" for both datasets but does not explicitly mention or detail a validation split or its size/percentage. |
| Hardware Specification | Yes | We conducted our experiments utilizing the Py Torch framework on four Titan XP GPUs |
| Software Dependencies | No | We conducted our experiments utilizing the Py Torch framework. While PyTorch is mentioned, a specific version number is not provided, nor are other key software dependencies with their versions. |
| Experiment Setup | Yes | Employing the Adam optimizer, we initially set the learning rate at 2 × 10^-4, progressively halving it at every 1 × 10^4 iterations to fine-tune the training process. |