Research map / Image generation / Text-to-image diffusion
CHIMERA: Controllable High-quality Image-Mask Extraction for Reliable Diffusion-based Anomaly Synthesis
JoungBin Lee, Hyunkoo Lee, Jini Yang, Chaehyun Kim, Jung Yi, Seok Hwangbo, Hyeoncheol Lee, Minho Chun, Eunjo Jeong, Seungryong Kim
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 focal mask loss on attention maps (ablation) | LoRA-only fine-tuning · MVTec AD ablation | PRO: 47.6 -> 90.4 | Improved |
| add focal mask loss on attention maps (ablation) | LoRA-only fine-tuning · MVTec AD ablation | AUC-P: 74.4 -> 96.8 | Improved |
| add mask decoder on top of mask loss (ablation) | LoRA + mask loss · MVTec AD ablation | F1-P: 64.9 -> 71.2 | Improved |
| add layer selection on top of decoder (ablation) | LoRA + mask loss + decoder · MVTec AD ablation | PRO: 90.4 -> 88.6 | No clear effect |
| use generic 'sks' token instead of anomaly-type text (ablation) | CHIMERA · MVTec AD ablation | PRO: 96.1 -> 37.4 | Hurt |
| omit spatial location from prompt (ablation) | CHIMERA · MVTec AD ablation | PRO: 96.1 -> 81.6 | Hurt |
| omit spatial location from prompt (ablation) | CHIMERA · MVTec AD ablation | F1-P: 75.0 -> 54.1 | Hurt |
| use structured VLM captions with SD3 and LoRA vs prior generators | SD3 CHIMERA · MVTec AD generation | Inception Score: 2.02 -> 2.40 | Improved |
| use CHIMERA vs AnomalyAnything | AnomalyAnything · MVTec AD generation | IC-LPIPS: 0.33 -> 0.27 | Hurt |
| use CHIMERA unified model vs per-class DRAEM | DRAEM · MVTec AD 1-shot detection | AUC-I: 94.6 -> 94.61 | No clear effect |
| use unified CHIMERA vs per-class AnoDiff | AnoDiff · MVTec AD 1-shot detection | AUC-I: 99.2 -> 94.61 | Hurt |
| use CHIMERA on unseen mobile panel data vs AnomalyAnything | AnomalyAnything · Mobile device panel 1-shot | AUC-I: 80.7 -> 84.1 | Improved |
Closest papers by meaning
- OmiAD: One-Step Adaptive Masked Diffusion Model for Multi-class Anomaly Detection via Adversarial Distillation (ICML 2025 · similarity 0.89)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.88)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.87)
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.87)
- Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · similarity 0.86)
- MMAD: A Comprehensive Benchmark for Multimodal Large Language Models in Industrial Anomaly Detection (ICLR 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.
