Research map / Detection and segmentation / CNN features
Mitigating Intensity Bias in Shadow Detection via Feature Decomposition and Reweighting
Lei Zhu, Ke Xu, Zhanghan Ke, Rynson W.H. Lau
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 the intensity-invariant consistency loss Li (ablation) | Basic+BB · SBU | BER: 3.32 -> 3.24 | Improved |
| add the intensity-variant prediction loss Lv alone (ablation) | Basic+BB · SBU | BER: 3.32 -> 3.36 | Hurt |
| add Lv to Basic+BB+Li (paired self-supervision) (ablation) | Basic+BB+Li · SBU | BER: 3.24 -> 3.04 | Improved |
| add bilateral branches without decomposition losses (ablation) | Basic · SBU | BER: 3.32 -> 3.32 | No clear effect |
| use ill-posed Lv (no reference image) instead of well-posed Lv (ablation) | FDRNet · SBU | BER: 3.21 -> 3.04 | Improved |
| replace constant mu=0.5 with parabolic decay cumulative learning (ablation) | FDRNet · SBU | BER: 3.45 -> 3.04 | Improved |
| replace learnable mu (backward) with parabolic decay (ablation) | FDRNet · SBU | BER: 3.47 -> 3.04 | Improved |
| replace mu predicted by extra branch (forward) with parabolic decay (ablation) | FDRNet · SBU | BER: 3.42 -> 3.04 | Improved |
| use decay instead of increment of mu (ablation) | FDRNet · SBU | BER: 3.63 -> 3.26 | Improved |
| use parabolic decay instead of linear decay of mu (ablation) | FDRNet · SBU | BER: 3.26 -> 3.04 | Improved |
| set test-time mu to extreme values 0 or 1 instead of 0.4 (ablation) | FDRNet · SBU | BER | Hurt |
| random brightness shift data augmentation (ablation) | FPN-like network · SBU | BER | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EGNet — EGNet: Edge Guidance Network for Salient Object Detection (ICCV 2019) · better on 3 of 3 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
- Unveiling Deep Shadows: A Survey and Benchmark on Image and Video Shadow Detection, Removal, and Generation in the Deep Learning Era (IJCV 2026 · similarity 0.86)
- Bidirectional Feature Pyramid Network with Recurrent Attention Residual Modules for Shadow Detection (ECCV 2018 · similarity 0.84)
- Instance Shadow Detection with A Single-Stage Detector (TPAMI 2022 · similarity 0.84)
- Direction-Aware Spatial Context Features for Shadow Detection (CVPR 2018 · similarity 0.84)
- Single-Stage Instance Shadow Detection With Bidirectional Relation Learning (CVPR 2021 · similarity 0.84)
- Direction-Aware Spatial Context Features for Shadow Detection and Removal (TPAMI 2019 · 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.
