Research map / Detection and segmentation / Weakly supervised detection and segmentation
Learning with Noisy Class Labels for Instance Segmentation
Longrong Yang, Fanman Meng, Hongliang Li, Qingbo Wu, Qishang Cheng
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 stage-wise training (CE early, RCE later) (ablation) | CE baseline · Pascal VOC (symmetric noise 40%) | AP: 31.5 -> 34.3 | Improved |
| add CE on negative and pseudo negative samples with RCE on the rest (ablation) | CE baseline · Pascal VOC (symmetric noise 40%) | AP: 31.5 -> 37.3 | Improved |
| add stage-wise training on top of N and PSN (ablation) | N+PSN · Pascal VOC (symmetric noise 40%) | AP: 37.3 -> 37.9 | Improved |
| add isolation of potential noisy samples (PON) (ablation) | ST+N+PSN · Pascal VOC (symmetric noise 40%) | AP: 37.9 -> 38.1 | Improved |
| add pseudo negative samples CE on top of ST and N (ablation) | ST+N · Pascal VOC (symmetric noise 40%) | AP: 37.4 -> 37.9 | Improved |
| add PON without stage-wise training (ablation) | N+PSN · Pascal VOC (symmetric noise 40%) | AP: 37.3 -> 36.5 | Hurt |
| first stage 9 epochs instead of 6 (E=12) (ablation) | Full method · Pascal VOC (symmetric noise 40%) | AP: 38.1 -> 34.6 | Hurt |
| first stage 3 epochs instead of 6 (E=12) (ablation) | Full method · Pascal VOC (symmetric noise 40%) | AP: 38.1 -> 37.2 | Hurt |
| gamma 4.0 instead of 6.0 (ablation) | Full method · Pascal VOC (symmetric noise 40%) | AP: 38.1 -> 37.5 | Hurt |
| gamma 8.0 instead of 6.0 (ablation) | Full method · Pascal VOC (symmetric noise 40%) | AP: 38.1 -> 37.2 | Hurt |
| gamma 5.0 or 7.0 instead of 6.0 (ablation) | Full method · Pascal VOC (symmetric noise 40%) | AP: 38.1 -> 38.0 | No clear effect |
| first stage 9 of 18 epochs instead of 6 of 18 (ablation) | Full method · Pascal VOC (symmetric noise 40%) | AP: 36.2 -> 38.5 | Improved |
Closest papers by meaning
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · similarity 0.89)
- Boosting Class Representation via Semantically Related Instances for Robust Long-Tailed Learning with Noisy Labels (ICCV 2025 · similarity 0.86)
- Part-dependent Label Noise: Towards Instance-dependent Label Noise (NeurIPS 2020 · similarity 0.86)
- Understanding Instance-Level Label Noise: Disparate Impacts and Treatments (ICML 2021 · similarity 0.86)
- Weakly Supervised Instance Segmentation by Learning Annotation Consistent Instances (ECCV 2020 · similarity 0.86)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 2025 · similarity 0.85)
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.
