Research map / Detection and segmentation / Open-set and few-shot detection
SALAD -- Semantics-Aware Logical Anomaly Detection
Matic Fučka, Vitjan Zavrtanik, Danijel Skočaj
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 a discriminative composition branch trained on composition maps (ablation) | SALAD · MVTec LOCO | image AUROC (logical): 93.0 -> 96.5 | Improved |
| add the appearance branch (ablation) | SALAD · MVTec LOCO | image AUROC (structural): 91.2 -> 95.7 | Improved |
| add the global (stat) branch (ablation) | SALAD · MVTec LOCO | image AUROC (avg): 94.2 -> 96.1 | Improved |
| use the three branches together instead of any single branch (ablation) | SALAD · MVTec LOCO | image AUROC (avg): 85.4 -> 96.1 (composition only vs full) | Improved |
| use class-wise global descriptor with spatially-grouped means instead of single global mean vector (ablation) | SALAD global branch · MVTec LOCO | image AUROC (logical): 95.6 -> 96.5 | Improved |
| use EfficientAD features rather than ResNet50 for global representation (ablation) | SALAD global branch · MVTec LOCO | image AUROC (avg): 92.7 -> 96.1 | Improved |
| use removal, inpainting and DRAEM synthetic anomalies together (ablation) | composition branch · MVTec LOCO | image AUROC (avg): 95.2 -> 96.1 (without removal) | Improved |
| swap appearance branch to TransFusion, DSR or DRAEM (ablation) | SALAD · MVTec LOCO | image AUROC (avg): 96.1 -> 95.4 (TransFusion), 94.4 (DSR), 94.0 (DRAEM) | Mixed |
| use DINOv2 instead of DINO for composition map generation (ablation) | SALAD · MVTec LOCO | image AUROC (avg): 96.1 -> 95.5 | Hurt |
| increase number of clusters from 4 to 6 or 8 (ablation) | composition map generation · MVTec LOCO | image AUROC (avg): 95.4 (K=4), 96.1 (K=6), 95.9 (K=8) | No clear effect |
| train SALAD with supervised PSAD composition maps | SALAD · MVTec LOCO | image AUROC (avg): 96.1 -> 97.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- TransFusion — TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers (CVPR 2022) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Few Shot Part Segmentation Reveals Compositional Logic for Industrial Anomaly Detection (AAAI 2024 · similarity 0.88)
- LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction (AAAI 2025 · similarity 0.86)
- Towards Training-free Anomaly Detection with Vision and Language Foundation Models (CVPR 2025 · similarity 0.85)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.85)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (CVPR 2025 · similarity 0.84)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.84)
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.
