For my master's thesis, We developed an AI-powered web application for industrial machine diagnosis and defect detection. Built with Next.js, it leverages advanced vision-language models to analyze images, detect anomalies, and provide real-time expert guidance for industrial quality control.
-
Prompt-based Defect Detection:
Use zero-shot vision-language models to detect unseen defects by describing them in text prompts. -
FiLo Heatmap Enhancement:
Localize defects with FiLo’s refined heatmaps and improve inference through preprocessed focus regions. -
Visual Question Answering (VQA):
Ask visual questions (e.g., "Is there a scratch on the metal part?") and get instant AI-driven answers. -
Interactive Pipeline:
Upload images, preprocess, overlay, and analyze them step-by-step with real-time feedback and markdown result streaming. -
Modern UI:
Responsive, animated interface using Tailwind CSS, Radix UI, and motion for smooth user experience.
- Node.js (v18+ recommended)
- npm, yarn, pnpm, or bun