lizuju/garbage-classification
🚀 基于 YOLOv5 的全栈垃圾分类识别系统。集成 Vue 3 玻璃拟态 UI、Flask 后端、JWT 权限体系及完整管理员后台。支持模型迭代训练、数据统计、日志审计与历史记录管理。
What's novel
🚀 基于 YOLOv5 的全栈垃圾分类识别系统。集成 Vue 3 玻璃拟态 UI、Flask 后端、JWT 权限体系及完整管理员后台。支持模型迭代训练、数据统计、日志审计与历史记录管理。
Code Analysis
19 files read · 4 roundsA full-stack garbage classification web app that uses a modified YOLOv5x-cls model (with a dual-head hierarchical loss for 265 fine-grained classes + 4 major categories) to classify uploaded images, wrapped in a Flask REST API with JWT auth, user management, admin panel, and a Vue 3 frontend with AM
Strengths
The dual-head hierarchical training modification to YOLOv5's classification pipeline is a genuine (if modest) ML contribution — the training loop, validation, and model architecture were all coherently modified to support joint fine/major-class optimization. The full-stack integration (Flask + Vue 3 + AMap + real-time frame averaging) is complete and functional.
Weaknesses
Zero test coverage across the entire project. The codebase contains two full copies of YOLOv5 (v5 and v6.2) adding ~260 files of vendored dependency code that dilutes the project's own contribution. The 'real-time' detection is just a 3-frame probability average with no actual video stream processing, and the class-name mapping logic (ImageNet dict-order → Chinese names) is fragile and hard to maintain.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
40
Contributors
1
Files
383
Active weeks
5
Repository
Language
Python
Stars
1
Forks
0
License
MIT