Research map / Detection and segmentation / Open-set and few-shot detection
Hybrid Open-Set Segmentation With Synthetic Negative Data
Matej Grcić, Siniša Šegvić
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 generative score to discriminative score (hybrid) (ablation) | DenseHybrid (real negatives) · Fishyscapes val LAF | AP: 46.5 -> 60.5 | Improved |
| add discriminative score to generative score (hybrid) (ablation) | DenseHybrid (real negatives) · Fishyscapes val LAF | AP: 58.2 -> 60.5 | Improved |
| add generative score to discriminative score (hybrid) (ablation) | DenseHybrid (synthetic negatives) · Fishyscapes val Static | AP: 48.8 -> 52.1 | Improved |
| add hybrid fusion over single detectors (ablation) | DenseHybrid · Pascal-COCO | open-mIoU: 35.8 -> 41.0 | Improved |
| add hybrid fusion over single detectors (ablation) | DenseHybrid · COCO 20/80 | open-mIoU: 15.2 -> 16.3 | Improved |
| replace pointwise normalizing-flow likelihood with unnormalized density LSE(s) (ablation) | DenseHybrid · Fishyscapes val LAF | AP: 5.7 -> 58.2 | Improved |
| replace flow likelihood with unnormalized density (ablation) | DenseHybrid · Fishyscapes val Static | AP: 58.0 -> 61.7 | Hurt |
| use uniform-classification JSD loss for the flow instead of L_d + L_x^UB (ablation) | normalizing flow negatives · Fishyscapes val LAF | AP: 46.1 -> 60.2 | Improved |
| use uniform-classification JSD loss for the flow (ablation) | normalizing flow negatives · Fishyscapes val Static | AP: 41.8 -> 52.1 | Improved |
| replace uniform-noise negatives with jointly trained flow negatives (ablation) | DenseHybrid · average AP over 4 sets | AP: 42.1 -> 67.5 | Improved |
| replace synthetic flow negatives with real ADE20k-instances (ablation) | DenseHybrid · average AP over 4 sets | AP: 67.5 -> 73.6 | Improved |
| use real ADE20k negatives instead of synthetic (ablation) | DenseHybrid · COCO 20/80 | open-mIoU: 16.3 -> 14.7 | Hurt |
Closest papers by meaning
- DenseHybrid: Hybrid Anomaly Detection for Dense Open-Set Recognition (ECCV 2022 · similarity 0.92)
- A Dataset for Semantic Segmentation in the Presence of Unknowns (CVPR 2025 · similarity 0.85)
- Anomaly Heterogeneity Learning for Open-set Supervised Anomaly Detection (CVPR 2024 · similarity 0.85)
- MetaUAS: Universal Anomaly Segmentation with One-Prompt Meta-Learning (NeurIPS 2024 · similarity 0.84)
- SetAD: Semi-Supervised Anomaly Learning in Contextual Sets (NeurIPS 2026 · similarity 0.84)
- Revisiting Open-Set Panoptic Segmentation (AAAI 2024 · similarity 0.83)
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.
