Research map / Detection and segmentation / Weakly supervised detection and segmentation
Modeling the Background for Incremental and Weakly-Supervised Semantic Segmentation
Fabio Cermelli, Massimiliano Mancini, Samuel Rota Bulò, Elisa Ricci, Barbara Caputo B
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 modified cross-entropy that sums old-class and background probability (ablation) | LwF · VOC 15-1 overlapped | mIoU all: 1.8 -> 9.9 | Improved |
| add modified distillation loss (KD) (ablation) | LwF + CE · VOC 15-1 overlapped | mIoU all: 9.9 -> 27.2 | Improved |
| add classifier initialization from background weights (ablation) | LwF + CE + KD · VOC 15-1 overlapped | mIoU all: 27.2 -> 29.7 | Improved |
| add classifier initialization (ablation) | LwF + CE + KD · VOC 19-1 overlapped | mIoU novel class: 11.9 -> 22.1 | Improved |
| add modified distillation loss (KD) (ablation) | LwF + CE · VOC 19-1 overlapped | mIoU all: 55.2 -> 63.3 | Improved |
| add modified KD (ablation) | LwF + CE · VOC 15-5 overlapped | mIoU all: 57.0 -> 66.3 | Improved |
| add classifier initialization (ablation) | LwF + CE + KD · VOC 15-5 overlapped | mIoU all: 66.3 -> 69.0 | Improved |
| replace standard LwF with background-aware MiB | LwF · VOC 15-1 overlapped | mIoU all: 1.8 -> 29.7 | Improved |
| model unlabeled pixels as background-or-annotated-classes instead of only background (ablation) | PCE point supervision · VOC point supervision | mIoU: 38.8 -> 46.7 | Improved |
| add unlabeled-pixel loss to partial cross-entropy | PCE · VOC scribble supervision (w/o CRF) | mIoU: 69.5 -> 72.3 | Improved |
| add unlabeled-pixel loss to partial cross-entropy | PCE · VOC scribble supervision (w/ CRF) | mIoU: 72.8 -> 75.1 | Improved |
| add unlabeled-pixel loss to partial cross-entropy | PCE · ADE20K point supervision (our protocol) | mIoU: 22.4 -> 22.9 | Improved |
Closest papers by meaning
- Modeling the Background for Incremental Learning in Semantic Segmentation (CVPR 2020 · similarity 0.95)
- Incremental Learning in Semantic Segmentation From Image Labels (CVPR 2022 · similarity 0.92)
- Mitigating Background Shift in Class-Incremental Semantic Segmentation (ECCV 2024 · similarity 0.92)
- Incrementer: Transformer for Class-Incremental Semantic Segmentation With Knowledge Distillation Focusing on Old Class (CVPR 2023 · similarity 0.91)
- Background Adaptation with Residual Modeling for Exemplar-Free Class-Incremental Semantic Segmentation (ECCV 2024 · similarity 0.91)
- Reliability Does Matter: An End-to-End Weakly Supervised Semantic Segmentation Approach (AAAI 2020 · 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.
