PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
March 29, 2026Algorithms6 citationsOpen Access

Adapting Vision–Language Models for Few-Shot Industrial Defect Detection

View Full Paper
CSChayanon Sub-r-paRCRung-Ching Chen

Key Points

  • This research aims to enhance automated surface defect detection in few-shot settings using Vision-Language Models.
  • Adapted Vision-Language Models (VLMs) using YOLO-World.
  • Applied semantic pre-training to reduce reliance on annotated data.
  • Evaluated the approach on the MVTec AD dataset with a strict train-validation split.
  • Performed ablation studies to assess the impact of different optimization techniques.
  • YOLO-World surpassed traditional object detectors in 12 of 15 categories.
  • Achieved an average mean Average Precision (mAP@50) of 64.9% for texture-heavy categories.
  • Identified critical factors affecting performance, including the necessity of disabling spatial distortions.
  • Showed that normalization techniques improved localization of defects.

Abstract

Automated surface defect detection often faces a “cold-start” problem due to limited annotated data for new anomalies. Traditional object detectors struggle to converge in such few-shot settings. To address this, we adapt Vision–Language Models (VLMs), specifically YOLO-World. We use semantic pre-training to mitigate data scarcity. We evaluate this approach on the MVTec AD dataset in bounding-box format. We use a strict 1:9 train-validation split, resulting in an average of 11.8 defect instances per category. YOLO-World surpasses traditional baselines, like YOLOv11s and YOLOv26s, in 12 of 15 categories. The optimized VLM pipeline achieves up to 64.9% mAP@50 on texture-heavy categories, such as Tile, with only nine training instances. Ablation studies show standard optimization techniques are limited under 10-shot constraints. We find a critical augmentation divide. Disabling spatial distortions (Mosaic) is vital to preserving rigid-object geometry. The Normalized Wasserstein Distance (NWD) improves the localization of microscopic anomalies. Varifocal Loss (VFL) often causes model collapse. Ultimately, VLMs offer a superior foundation for cold-start inspection but require carefully tailored pipelines for robustness.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Sub-r-pa et al. (2026) studied this question.

synapsesocial.com/papers/69c8c277de0f0f753b39cb72https://doi.org/10.3390/a19040259
Ask AI
Helpful
Bookmark
Share
View Full Paper