Research map / Image generation / Image restoration
Context-Aware Image Matting for Simultaneous Foreground and Alpha Estimation
Qiqi Hou, Feng Liu
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 context encoder (ablation) | ME + L_lap · Composition-1K (alpha) | SAD: 43.9 -> 35.8 | Improved |
| make matting-encoder-only model deeper (117.0M params) instead of adding context encoder (ablation) | ME (model 2) · Composition-1K (alpha) | SAD: 43.9 -> 43.7 | No clear effect |
| replace Deep Matting loss with Laplacian loss (ablation) | ME · Composition-1K (alpha) | SAD: 49.1 -> 43.9 | Improved |
| add feature loss (ablation) | ME + CE + L_lap · Composition-1K (alpha) | SAD: 35.8 -> 38.8 | Hurt |
| add feature loss (ablation) | ME + CE + L_lap · Real-world images (NIMA score) | NIMA mean score: 4.64 -> 4.69 | No clear effect |
| add foreground decoder color losses (L1c, LFc) (ablation) | ME + CE + L_lap + L_F · Composition-1K (alpha) | SAD: 38.8 -> 38.0 | Improved |
| add foreground decoder color losses (L1c, LFc) with DA (ablation) | ME + CE + L_lap + L_F + DA · Composition-1K (alpha) | SAD: 71.3 -> 84.1 | Hurt |
| add data augmentation (ablation) | ME + CE + L_lap + L_F · Composition-1K (alpha) | SAD: 38.8 -> 71.3 | Hurt |
| add data augmentation (ablation) | ME + CE + L_lap + L_F · Real-world images (NIMA score) | NIMA mean score: 4.69 -> 5.03 | Improved |
| add ReJPEGing augmentation (ablation) | model 7 without ReJPEGing · Composition-1K (alpha) | SAD: 55.1 -> 84.1 | Hurt |
| add Gaussian blur augmentation (ablation) | model 7 without GaussianBlur · Composition-1K (alpha) | SAD: 69.1 -> 84.1 | Hurt |
| add data augmentation (ablation) | ME + CE + L_lap + L_F + color losses · Composition-1K (foreground) | SAD: 61.72 -> 94.41 | Hurt |
Closest papers by meaning
- Deep Image Matting (CVPR 2017 · similarity 0.93)
- Bridging Composite and Real: Towards End-to-End Deep Image Matting (IJCV 2022 · similarity 0.93)
- In-Context Matting (CVPR 2024 · similarity 0.92)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.91)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.91)
- DRIP: Unleashing Diffusion Priors for Joint Foreground and Alpha Prediction in Image Matting (NeurIPS 2024 · similarity 0.91)
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.
