Research map / Detection and segmentation / Open-set and few-shot detection
Pixel-Wise Energy-Biased Abstention Learning for Anomaly Segmentation on Complex Urban Driving Scenes
Yu Tian, Yuyuan Liu, Guansong Pang, Fengbei Liu, Yuanhong Chen, Gustavo Carneiro
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 PAL loss with energy-biased penalty to EBM (ablation) | PEBAL · LostAndFound test (WideResnet38) | AP: 70.24 -> 77.19 | Improved |
| add PAL loss with energy-biased penalty to EBM (ablation) | PEBAL · LostAndFound test (WideResnet38) | FPR95: 8.92 -> 1.19 | Improved |
| add EBM with adaptive penalty to fixed-penalty PAL (ablation) | PEBAL · LostAndFound test (WideResnet38) | AP: 72.73 -> 77.19 | Improved |
| replace entropy-maximisation baseline with joint EBM + PAL (ablation) | PEBAL · LostAndFound test (WideResnet38) | AP: 69.02 -> 77.19 | Improved |
| add smoothness and sparsity regularisation (ablation) | PEBAL · LostAndFound test (WideResnet38) | AP: 77.19 -> 78.29 | Improved |
| add smoothness and sparsity regularisation (ablation) | PEBAL · LostAndFound test (WideResnet38) | FPR95: 1.19 -> 0.81 | Improved |
| increase share of COCO outlier classes from 1% to 75% (ablation) | PEBAL · Fishyscapes LostAndFound val | AP: 53.57 -> 57.86 | No clear effect |
| increase share of COCO outlier classes from 1% to 75% (ablation) | PEBAL · Fishyscapes Static val | AP: 85.84 -> 91.85 | Improved |
| increase outlier training set size from 5% to 50% (ablation) | PEBAL · Fishyscapes LostAndFound val | AP: 54.32 -> 57.34 | No clear effect |
| increase outlier training set size from 5% to 50% (ablation) | PEBAL · Fishyscapes Static val | AP: 89.11 -> 91.29 | Improved |
| set lambda to 0.01 instead of 0.1 (ablation) | PEBAL · LostAndFound test (WideResnet38) | AP: 77.15 -> 78.29 | No clear effect |
| fine-tune only the final classification block instead of re-training whole network | PEBAL · trainable parameters | parameters: 137.1M -> 1.3M | Improved |
Closest papers by meaning
- Self-Supervised Likelihood Estimation with Energy Guidance for Anomaly Segmentation in Urban Scenes (AAAI 2024 · similarity 0.90)
- SDAC: A Multimodal Synthetic Dataset for Anomaly and Corner Case Detection in Autonomous Driving (AAAI 2024 · similarity 0.85)
- Spotting the Unexpected (STU): A 3D LiDAR Dataset for Anomaly Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.84)
- FAST: Foreground‑aware Diffusion with Accelerated Sampling Trajectory for Segmentation‑oriented Anomaly Synthesis (NeurIPS 2025 · similarity 0.84)
- Promptable Anomaly Segmentation with SAM Through Self-Perception Tuning (AAAI 2025 · similarity 0.83)
- When Every Millisecond Counts: Real-Time Anomaly Detection via the Multimodal Asynchronous Hybrid Network (ICML 2025 · 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.
