Object Detection for classification of recycled items

📘Overview

2024.10.02 ~ 2024.10.24

This project focuses on detecting objects in recycling trash images as part of a private competition organized by Naver Connect Foundation and Upstage.

📘Contributors

김기수	문채원	안주형	은의찬	이재훈	장지우

📘Wrap up Report

Here's our link

📘Metrics

• mAP50

📰Tools

• github
• notion
• slack
• wandb

📰Folder Structure

├── mmdetection2
│   ├── configs
│   ├──projects/configs/custom
│   ├── inference_wbf.py
│   ├── train.py
│   └── etc
├── mmdetection3
│   └── config files
├── Ultralytics
│   ├── RT_DETR.ipynb
│   ├── RT_DETR_WBF_infernece.ipynb
│   ├── Yolov11.ipynb
│   └── Yolov11_WBF_infernece.ipynb
├── data_split.py
├── eda.ipynb
├── visualization_data.ipynb
└── wbf.py

📰Dataset Structure

├── dataset
    ├── train.json
    ├── test.json
    ├── train
    └── test

• images : 9754
  • train : 4883
  • test : 4871
• 10 class : General trash, Paper, Paper pack, Metal, Glass, Plastic, Styrofoam, Plastic bag, Battery, Clothing
• image size : (1024, 1024)

📰Models

• Faster-RCNN
• Cascade-RCNN
• Atss
• YOLOs(3, 5, 9, x)
• Co-Deformable-DETR
• RetinaNet
• YOLOv11
• Co-DINO
• RT-DETR

📰Backbones

• Swin Transformer
• Resnet

📰Experiments

Exp	mAP
Yolov11(5), RT-DETR(5), CO-DINO(2)	0.6760
Co-dino_r50(2), Co-dino_swin(5)	0.6590
Co-dino_swin(5),Co-dino_r50(2),RT-DETR(5)	0.6797
Co_dino_swin(5), RT-DETR(5), Yolov11(5)	0.6834