Research map / Image generation / Image restoration
CIE XYZ Net: Unprocessing Images for Low-Level Computer Vision Tasks
Mahmoud Afifi, Abdelrahman Abdelhamed, Abdullah Abuolaim, Abhijith Punnappurath, Michael S. Brown
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn CIE XYZ decomposition instead of standard gamma mapping | sRGB to XYZ · MIT-Adobe FiveK test | PSNR (dB): 21.84 -> 29.66 | Improved |
| use CIE XYZ Net instead of UPI | sRGB to XYZ · MIT-Adobe FiveK test | PSNR (dB): 22.19 -> 29.66 | Improved |
| use two-part global plus residual local network instead of U-Net (ablation) | sRGB to XYZ · MIT-Adobe FiveK test | PSNR (dB): 20.05 -> 29.66 | Improved |
| use two-part network instead of U-Net (ablation) | Rec. XYZ to sRGB · MIT-Adobe FiveK test | PSNR (dB): 43.39 -> 43.82 | Improved |
| denoise CBM3D in our XYZ instead of standard XYZ | CBM3D · SIDD-Validation | PSNR (dB): 33.84 -> 35.49 | Improved |
| denoise CBM3D in our XYZ instead of sRGB | CBM3D · SIDD-Validation | PSNR (dB): 34.44 -> 35.49 | Improved |
| denoise DnCNN-P in our XYZ instead of standard XYZ | DnCNN-P · SIDD-Validation | PSNR (dB): 26.23 -> 27.71 | Improved |
| use our XYZ instead of standard XYZ for retrained DnCNN | DnCNN-R · SIDD-Validation | PSNR (dB): 33.92 -> 33.00 | Hurt |
| use our XYZ instead of standard XYZ for retrained DnCNN | DnCNN-R · SIDD-Validation | SSIM: 0.7833 -> 0.8865 | Improved |
| deblur in our XYZ space instead of sRGB | Krishnan and Fergus deblurring · FiveK 244 images | PSNR (dB): 29.41 -> 30.26 | Improved |
| deblur in our XYZ space instead of standard XYZ | Krishnan and Fergus deblurring · FiveK 244 images | PSNR (dB): 28.68 -> 30.26 | Improved |
| add our reconstructed raw-like images to real training data | illuminant estimation · NUS Canon EOS-1Ds Mark III | mean angular error: 4.15 -> 2.41 | Improved |
Later papers that compared with this method
1 later accepted paper reports CIE XYZ Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- RawGen: Learning Camera Raw Image Generation (ECCV 2026 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Unprocessing Images for Learned Raw Denoising (CVPR 2019 · similarity 0.85)
- Model-Based Image Signal Processors via Learnable Dictionaries (AAAI 2022 · similarity 0.85)
- Graphics2RAW: Mapping Computer Graphics Images to Sensor RAW Images (ICCV 2023 · similarity 0.84)
- ReRAW: RGB-to-RAW Image Reconstruction via Stratified Sampling for Efficient Object Detection on the Edge (CVPR 2025 · similarity 0.84)
- Logarithmic Lenses: Exploring Log RGB Data for Image Classification (CVPR 2024 · similarity 0.84)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
