Research map / Graph learning / Knowledge and temporal graphs
Cascade Image Matting With Deformable Graph Refinement
Zijian Yu, Xuhui Li, Huijuan Huang, Wen Zheng, Li Chen
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 network cascade (5 stages instead of 1) (ablation) | matting network · Adobe human matting test | SAD: 3.78 -> 2.92 | Improved |
| add network cascade (5 stages instead of 1) (ablation) | matting network · Adobe human matting test | Grad: 4.67 -> 2.85 | Improved |
| add DGR refinement (K=5, 2 iterations) to cascade (ablation) | Ours-Cascade · Adobe human matting test | SAD: 2.92 -> 1.76 | Improved |
| add DGR refinement (K=5, 2 iterations) to cascade (ablation) | Ours-Cascade · Adobe human matting test | Grad: 2.85 -> 1.66 | Improved |
| increase DGR neighbors from 1 to 5 (1 iteration) (ablation) | CasDGR · Adobe human matting test | SAD: 2.25 -> 1.93 | Improved |
| increase DGR neighbors from 5 to 9 (1 iteration) (ablation) | CasDGR · Adobe human matting test | SAD: 1.93 -> 2.16 | Hurt |
| increase DGR neighbors from 5 to 9 (2 iterations) (ablation) | CasDGR · Adobe human matting test | SAD: 1.76 -> 1.84 | Hurt |
| increase DGR refinement iterations from 1 to 2 (K=5) (ablation) | CasDGR · Adobe human matting test | SAD: 1.93 -> 1.76 | Improved |
| increase DGR refinement iterations from 1 to 2 (K=1) (ablation) | CasDGR · Adobe human matting test | SAD: 2.25 -> 2.05 | Improved |
| increase DGR refinement iterations from 1 to 2 (K=5) (ablation) | CasDGR · Adobe human matting test (512x512, RTX 3090) | Inference time (ms): 41.33 -> 48.56 | Hurt |
| replace deformable convolution (DCN) with DGR (1 layer) (ablation) | Ours-Cascade · Adobe human matting test | SAD: 2.13 -> 1.93 | Improved |
| replace deformable convolution (DCN) with DGR (1 layer) (ablation) | Ours-Cascade · Adobe human matting test (512x512, RTX 3090) | Inference time (ms): 51.23 -> 41.33 | Improved |
Closest papers by meaning
- A Late Fusion CNN for Digital Matting (CVPR 2019 · similarity 0.90)
- Deep Image Matting (CVPR 2017 · similarity 0.90)
- Bridging Composite and Real: Towards End-to-End Deep Image Matting (IJCV 2022 · similarity 0.89)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.88)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.86)
- Path-Adaptive Matting for Efficient Inference Under Various Computational Cost Constraints (AAAI 2025 · similarity 0.86)
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.
