Image + VLM Anomaly Detection System
An anomaly detection system that combines image classification with a VLM to judge normal versus anomalous. Runs entirely within Docker.
Duration
Contract development
Team
1 person
01 // OVERVIEW
Project Overview
A system that detects anomalies from images. It judges using a combination of a classifier and a VLM (Vision Language Model), keeping the judgment logic and data management compact. The environment can be reproduced with Docker Compose.
02 // KEY FEATURES
Key Features
Combined Classifier and VLM
Built a judgment engine that combines image classification with VLM-based evaluation to determine whether an anomaly is present.
Fully Containerized Setup
The judgment engine, data layer, and VLM integration are all contained within Docker, enabling deployment with minimal environment dependency.
03 // TECHNICAL DEEP-DIVE
Technical Highlights
Solution
Division of Roles Between Classification and VLM
A lightweight classifier makes the initial judgment, and the VLM supplements cases that are difficult to judge, balancing accuracy against processing cost.
04 // IMPACT & RESULTS
Results
Classifier + VLM
Two-stage judgment
Docker-Contained
Minimal environment dependency