Research map / Image generation / Image restoration
Reconfiguring the Imaging Pipeline for Computer Vision
Mark Buckler, Suren Jayasuriya, Adrian Sampson
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 |
|---|---|---|---|
| add demosaicing to the pipeline (versus removing it) (ablation) | Farneback optical flow · Middlebury optical flow | mean error: 0.448 -> 0.227 | Improved |
| add gamma compression (versus removing it) (ablation) | OpenMVG SfM · Strecha | feature detection success: fails entirely without gamma compression | Improved |
| add the color transform stage (ablation) | OpenMVG SfM · Strecha | RMSE: 0.445 -> 0.408 | Improved |
| add gamut mapping (ablation) | ResNet44 · CIFAR-10 | top-1 error (%): 6.6% -> 6.3% | Improved |
| add sensor noise to the input | Stereo SGBM · Middlebury stereo | mean error: 0.245 -> 0.425 | Hurt |
| skip gamut mapping or gamma compression (ablation) | OpenFace · LFW/CASIA face identification | error (%): 8.65% -> 7.9% and 8.13% | No clear effect |
| add denoising to the minimal demosaic + gamma pipeline (ablation) | Stereo SGBM · Middlebury stereo | mean error: 0.33 -> 0.25 | Improved |
| use only demosaicing and gamma compression instead of the full ISP (ablation) | ResNet44 · CIFAR-10 | top-1 error (%): 6.3% -> 7.2% | Hurt |
| replace ISP demosaicing with in-sensor subsampling (ablation) | Farneback optical flow · Middlebury optical flow | mean error: 0.332 -> 0.375 | Hurt |
| switch from linear to logarithmic ADC (8 bits) (ablation) | Farneback optical flow · Middlebury optical flow | mean error: 0.54 -> 0.33 | Improved |
| switch from linear to logarithmic ADC (2 bits) (ablation) | Farneback optical flow · Middlebury optical flow | mean error: 0.65 -> 0.38 | Improved |
| reduce logarithmic ADC bitwidth from 8 to 5 (ablation) | OpenMVG SfM · Strecha | RMSE: 0.454 -> 0.474 | No clear effect |
Closest papers by meaning
- Hardware-in-the-Loop End-to-End Optimization of Camera Image Processing Pipelines (CVPR 2020 · similarity 0.88)
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection (NeurIPS 2024 · similarity 0.87)
- ReconfigISP: Reconfigurable Camera Image Processing Pipeline (ICCV 2021 · similarity 0.86)
- ASP Vision: Optically Computing the First Layer of Convolutional Neural Networks Using Angle Sensitive Pixels (CVPR 2016 · similarity 0.86)
- RL-SeqISP: Reinforcement Learning-Based Sequential Optimization for Image Signal Processing (AAAI 2024 · similarity 0.86)
- RGB-Event ISP: The Dataset and Benchmark (ICLR 2025 · 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.
