Research map / Image generation / Image restoration
Bridging Composite and Real: Towards End-to-End Deep Image Matting
Jizhizi Li, Jing Zhang, Stephen J. Maybank, Dacheng Tao
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 |
|---|---|---|---|
| use glance and focus decoders instead of a single decoder (ablation) | GFM-TT(d) · AM-2k ORI-Track | SAD: 13.79 -> 10.26 | Improved |
| use glance and focus decoders instead of a single decoder (ablation) | GFM-TT(r) · AM-2k ORI-Track | SAD: 15.50 -> 10.89 | Improved |
| add PPM (ablation) | GFM-TT(d) · AM-2k ORI-Track | SAD: 10.86 -> 10.26 | Improved |
| add bridge block (BB) (ablation) | GFM-TT(d) · AM-2k ORI-Track | SAD: 11.27 -> 10.26 | Improved |
| use two PPMs instead of one (ablation) | GFM-TT(d) · AM-2k ORI-Track | SAD: 10.71 -> 10.26 | Improved |
| modify ResNet-34 encoder E0 (r2b) (ablation) | GFM-TT(r) · AM-2k ORI-Track | SAD: 10.89 -> 10.24 | Improved |
| add large-aperture blur only (ablation) | GFM-TT(d) · AM-2k COMP-RSSN | SAD: 25.19 -> 21.37 | Improved |
| add noise on composite images only (ablation) | GFM-TT(d) · AM-2k COMP-RSSN | SAD: 25.19 -> 19.87 | Improved |
| use all RSSN techniques (ablation) | GFM-TT(d) · AM-2k COMP-RSSN | SAD: 25.19 -> 19.19 | Improved |
| train SHM with RSSN instead of BG-20K composition | SHM · AM-2k | SAD: 52.36 -> 23.94 | Improved |
| use RIM integration instead of TT alone | GFM · AM-2k | SAD: 11.28 -> 10.79 | Improved |
| ensemble by median instead of RIM | GFM · AM-2k | inference time: 0.1859 -> 0.4644 | Hurt |
Closest papers by meaning
- Context-Aware Image Matting for Simultaneous Foreground and Alpha Estimation (ICCV 2019 · similarity 0.93)
- Semantic Image Matting (CVPR 2021 · similarity 0.92)
- In-Context Matting (CVPR 2024 · similarity 0.91)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.90)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.90)
- Unifying Automatic and Interactive Matting with Pretrained ViTs (CVPR 2024 · similarity 0.90)
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.
