Unification of Closed-Open Industrial Detection Scenarios: New Large-Scale Benchmarks,Challenges and Baselines
Researchers have introduced a new large-scale benchmark, MMIOC-1M, designed to improve the application of Large-Scale Visual-Language Models (LVLMs) in industrial defect detection. This benchmark contains over one million samples across numerous defect categories and industrial scenes, aiming to provide extensive pre-training data for LVLMs in this domain. To address limitations in manual prompting and fine-grained understanding, they also propose RTVPNet, a model incorporating domain adaptation, automatic prompt generation, and enhanced text-visual interaction. AI
IMPACT Enhances LVLM capabilities for industrial applications, potentially improving quality control and reducing manufacturing defects.