Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
SFR-Net: Steering-Fusion-Refining Network in Multi-label Zero-Shot Sewer Defect Detection
Zhao-Min Chen, Xinjian Huang, Yisu Ge, Yu Li
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 Representation Steering (RS) module to CLIP (ablation) | CLIP ViT-B/16 · Sewer-ML ML-GZSL | mAP: 9.92% -> 39.62% | Improved |
| add Multi-Granularity Evidence Fusion (MEF) (ablation) | SFR-Net · Sewer-ML ML-ZSL | mAP: 8.36 -> 9.81 | Improved |
| add Generalized Relational Score Refining (GR) (ablation) | SFR-Net · Sewer-ML ML-ZSL | mAP: 9.81 -> 12.25 | Improved |
| add rank loss (ablation) | SFR-Net · Sewer-ML ML-GZSL | mAP: 41.54 -> 43.28 | Improved |
| use LLM-generated descriptions instead of base template (ablation) | SFR-Net · Sewer-ML ML-ZSL | mAP: 7.13 -> 12.58 | Improved |
| use manual short descriptions instead of base template (ablation) | SFR-Net · Sewer-ML ML-ZSL | mAP: 7.13 -> 6.90 | Hurt |
| fine-tune image encoder only vs RS on all layers (ablation) | RS module · Sewer-ML ML-ZSL | mAP: 8.36 -> 6.35 | Hurt |
| partial fine-tune (last three layers) vs all layers (ablation) | RS module · Sewer-ML | mAP ZSL/GZSL: ZSL 8.36 -> 10.15; GZSL 39.62 -> 35.05 | Mixed |
| replace semantic similarity adjacency with identity matrix (ablation) | GR module · Sewer-ML ML-ZSL | mAP: 12.58 -> 7.74 | Hurt |
| replace semantic adjacency with random matrix (ablation) | GR module · Sewer-ML ML-ZSL | mAP: 12.58 -> 8.28 | Hurt |
| use SFR-Net instead of RAM | CLIP ViT-B/16 · Sewer-ML ML-GZSL | mAP: 23.88 -> 43.28 | Improved |
Closest papers by meaning
- Sewer-ML: A Multi-Label Sewer Defect Classification Dataset and Benchmark (CVPR 2021 · similarity 0.89)
- MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context (AAAI 2025 · similarity 0.83)
- Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline (AAAI 2025 · similarity 0.83)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.81)
- Towards Open-Vocabulary Industrial Defect Understanding with a Large-Scale Multimodal Dataset (CVPR 2026 · similarity 0.81)
- Cracks in the Foundation: A Civil Infrastructure Dataset to Challenge Vision Foundation Models (NeurIPS 2026 · similarity 0.81)
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.
