Research map / Detection and segmentation / Open-set and few-shot detection
Semantic Visual Anomaly Detection and Reasoning in AI-Generated Images
Chuangchuang Tan, Xiang Ming, Jinglu Wang, Renshuai Tao, Bin Li, Yunchao Wei, Yao Zhao, Yan Lu
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 |
|---|---|---|---|
| LoRA fine-tune Qwen2.5-VL-7B on AnomReason train split | Qwen2.5-VL-7B · AnomReason-Test | SemAP (Full): 0.4155 -> 0.5162 | Improved |
| LoRA fine-tune Qwen2.5-VL-7B on AnomReason train split | Qwen2.5-VL-7B · AnomReason-Test | SemF1 (Full): 0.3775 -> 0.5009 | Improved |
| fine-tune on AnomReason versus proprietary GPT-4o | AnomReasonor-7B · AnomReason-Test | SemF1 (Full): 0.5109 (GPT-4o) vs 0.5009 | Mixed |
| fine-tune on AnomReason for explainable deepfake detection | Qwen2.5-VL-7B · AnomReason-Deepfake | Acc: 65.41 -> 82.61 | Improved |
| fine-tune on AnomReason for explanation quality | AnomReasonor-7B · AnomReason-Deepfake | CSemF1 (Rea): 0.3929 vs GPT-4o 0.3770 | Mixed |
| scale up Qwen2.5-VL from 7B to 72B (ablation) | Qwen2.5-VL · AnomReason-Test | SemAP (Full): 0.4155 -> 0.4568 | Improved |
| use reasoning vs observation view of anomalies (ablation) | off-the-shelf VLMs · AnomReason-Test | SemAP: InternVL3-8B 0.4552 -> 0.3676 | Hurt |
| add human-in-the-loop verification to AnomAgent annotations | AnomReason annotations · annotation pipeline | valid annotations per image: ~8 -> ~5.9 | Mixed |
Closest papers by meaning
- Anomagic: Crossmodal Prompt-driven Zero-shot Anomaly Generation (AAAI 2026 · similarity 0.88)
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.87)
- BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors (NeurIPS 2026 · similarity 0.86)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · similarity 0.86)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
- LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction (AAAI 2025 · similarity 0.86)
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.
