Research map / Video / Video segmentation and tracking
Two-Stage Video Shadow Detection via Temporal-Spatial Adaption
Xin Duan, Yu Cao, Lei Zhu, Gang Fu, Xin Wang, Renjie ZHANG, Ping Li
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 Temporal-Adaption and Spatial-Adaption blocks to the stage-1 image model (ablation) | Segformer shadow detector · ViSha | BER: 18.94 -> 8.96 | Improved |
| add both blocks on CVSD (ablation) | Segformer shadow detector · CVSD | BER: 24.58 -> 18.32 | Improved |
| pretrain an image-level encoder in stage 1 instead of training stage 2 from random initialization (ablation) | two-stage paradigm · ViSha | BER: 38.38 -> 18.94 (baseline) | Improved |
| directly transfer pretrained model by feeding references and patches through a shared encoder (ablation) | two-stage paradigm · ViSha | BER: 18.94 -> 23.83 | Hurt |
| add Spatial-Adaption Block alone (ablation) | Segformer shadow detector · CVSD | Fbeta: 0.445 -> 0.629 | Improved |
| add Temporal-Adaption Block alone (ablation) | Segformer shadow detector · CVSD | BER: 24.58 -> 19.03 | Improved |
| use 4 reference frames instead of 0 (ablation) | Segformer shadow detector · ViSha | BER: 17.63 -> 8.96 | Improved |
| use video methods versus image shadow detection methods | video shadow detection · ViSha and CVSD | overall | Improved |
| evaluate on complex CVSD dataset vs ViSha | all methods · CVSD | MAE / BER: Ours MAE 0.027 on ViSha vs 0.046 on CVSD | 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.
- SCOTCH and SODA — SCOTCH and SODA: A Transformer Video Shadow Detection Framework (CVPR 2023) · 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
- Triple-Cooperative Video Shadow Detection (CVPR 2021 · similarity 0.91)
- DTTNet: Improving Video Shadow Detection via Dark-Aware Guidance and Tokenized Temporal Modeling (AAAI 2026 · similarity 0.89)
- EDM:Event-guided Diffusion Model for Video Shadow Detection in Complex Dynamic Scenes (ECCV 2026 · similarity 0.88)
- Learning Shadow Correspondence for Video Shadow Detection (ECCV 2022 · 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)
- MetaShadow: Object-Centered Shadow Detection, Removal, and Synthesis (CVPR 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.
