Research map / Detection and segmentation / CNN features
Direction-Aware Spatial Context Features for Shadow Detection and Removal
Xiaowei Hu, Chi‐Wing Fu, Lei Zhu, Jing Qin, Pheng‐Ann Heng
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 spatial context (without direction-aware attention) to the basic network (ablation) | basic network · SBU shadow detection | BER: 6.55 -> 6.23 | Improved |
| add direction-aware attention weights to the spatial context (ablation) | basic+context · SBU shadow detection | BER: 6.23 -> 5.59 | Improved |
| use two rounds of recurrent translation instead of one (ablation) | DSC · SBU shadow detection | BER: 5.85 -> 5.59 | Improved |
| use three rounds instead of two with shared attention weights (ablation) | DSC · SBU shadow detection | BER: 5.59 -> 5.85 | Hurt |
| share attention weights across rounds instead of separate copies (two rounds) (ablation) | DSC · SBU shadow detection | BER: 6.02 -> 5.59 | Improved |
| replace VGG with ResNet-101 feature extractor (ablation) | DSC · SBU shadow detection | BER: 5.59 -> 5.73 | No clear effect |
| add CRF post-processing (ablation) | DSC · SBU shadow detection | BER: 5.68 -> 5.59 | Improved |
| train with color-adjusted shadow-free images (DSC+) and evaluate on adjusted ground truth (ablation) | DSC · ISTD shadow removal | RMSE: 8.54 -> 4.90 | Improved |
| train with color-adjusted shadow-free images (DSC+) and evaluate on adjusted ground truth (ablation) | DSC · SRD shadow removal | RMSE: 6.66 -> 6.12 | Improved |
| add color compensation, measured on non-shadow regions of ISTD against input (ablation) | DSC · ISTD shadow removal | RMSE: 7.07 -> 3.36 | Improved |
| use RGB instead of LAB color space (ablation) | DSC · SRD shadow removal | RMSE: 6.21 -> 6.05 | Mixed |
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.
- DeshadowNet — DeshadowNet: A Multi-Context Embedding Deep Network for Shadow Removal (CVPR 2017) · better on 1 of 1 shared result
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
- Direction-Aware Spatial Context Features for Shadow Detection (CVPR 2018 · similarity 0.96)
- Bidirectional Feature Pyramid Network with Recurrent Attention Residual Modules for Shadow Detection (ECCV 2018 · similarity 0.91)
- Efficient Model-Driven Network for Shadow Removal (AAAI 2022 · similarity 0.89)
- Automatic Shadow Detection and Removal from a Single Image (TPAMI 2016 · similarity 0.88)
- Unveiling Deep Shadows: A Survey and Benchmark on Image and Video Shadow Detection, Removal, and Generation in the Deep Learning Era (IJCV 2026 · similarity 0.88)
- DeshadowNet: A Multi-Context Embedding Deep Network for Shadow Removal (CVPR 2017 · similarity 0.88)
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.
