Research map / Detection and segmentation / CNN features
Instance Segmentation in the Dark
Linwei Chen, Ying Fu, Kaixuan Wei, Dezhi Zheng, Felix Heide
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 adaptive weighted downsampling (AWD) layer (ablation) | Mask R-CNN R50-FPN · LIS test | AP: 38.0 -> 39.3 | Improved |
| add smooth-oriented convolutional block (SCB) to AWD model (ablation) | Mask R-CNN R50-FPN · LIS test | AP: 39.3 -> 39.9 | Improved |
| add disturbance suppression learning (DSL) (ablation) | Mask R-CNN R50-FPN with AWD+SCB · LIS test (synthetic LIS) | AP: 39.9 -> 40.8 | Improved |
| use DSL instead of perceptual loss (ablation) | AWD+SCB baseline · LIS RAW-dark | AP: 39.9 -> 40.2 (perceptual loss) vs 39.9 -> 40.8 (DSL) | Improved |
| replace fixed Gaussian filter with learned AWD (ablation) | feature downsampling · LIS test | AP: 38.3 -> 39.3 | Improved |
| add AWD on top of CBAM (ablation) | Mask R-CNN with CBAM · LIS test | AP: 38.7 -> 40.0 | Improved |
| add global pooling branch to AWD (ablation) | AWD layer · LIS test | AP: 39.3 -> 39.1 | No clear effect |
| enlarge AWD kernel from 3x3 to 5x5 (ablation) | AWD layer · LIS test | AP: 39.3 -> 38.4 | Hurt |
| train on synthetic low-light RAW from sRGB images instead of direct sRGB | Mask R-CNN · LIS test | AP: 19.8 -> 31.8 | Improved |
| add enhancer Zero-DCE before Mask R-CNN | Mask R-CNN · LIS test | AP: 19.8 -> 19.7 | No clear effect |
| add denoiser SGN after Zero-DCE | Zero-DCE + Mask R-CNN · LIS test | AP: 19.7 -> 26.5 | Improved |
| add noise injection and unprocessing to synthetic data (ablation) | Mask R-CNN trained on LIS sRGB-normal · LIS RAW-dark | AP: 31.6 -> 38.0 | Improved |
Closest papers by meaning
- ELVIS: Enhance Low-Light for Video Instance Segmentation in the Dark (CVPR 2026 · similarity 0.86)
- Learning to See in the Dark (CVPR 2018 · similarity 0.86)
- SOLO: A Simple Framework for Instance Segmentation (TPAMI 2021 · similarity 0.86)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.85)
- PointINS: Point-Based Instance Segmentation (TPAMI 2021 · similarity 0.85)
- FRBNet: Revisiting Low-Light Vision through Frequency-Domain Radial Basis Network (NeurIPS 2025 · similarity 0.84)
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.
