Reproduce “VD-DeepStack: Bridging Visual Comparison and Language Reasoning for Few-Shot Anomaly Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mengyang Zhao, Zhuolin He, Haiyang Yu, Yuxuan Liang, Yifang Xu, Yuchuan Wu, Xiaolei Chen, Zhengtao Yao and 4 more
VD-DeepStack addresses reference-based visual inspection by supplying a model with spatially located visual differences alongside language reasoning. It matches query patches to reference patches, forms difference vectors, and injects them into decoder states. In one-shot evaluation across four industrial benchmarks, the 7B model averages 82.0% detection and 55.3% localization accuracy at IoU ≥ 0.1, exceeding Anomaly-R1-7B by 4.7 and 8.5 points.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.