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.
| 김기수 | 문채원 | 안주형 | 은의찬 | 이재훈 | 장지우 |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
- mAP50
- github
- notion
- slack
- wandb
├── baseline
│ ├── detectron2
│ │ ├── faster_rcnn_train.ipynb
│ │ ├── faster_rcnn_inference.ipynb
│ │ └── etc
│ ├── faster_rcnn
│ │ ├── faster_rcnn_torchvision_train.ipynb
│ │ └── faster_rcnn_torchvision_inference.ipynb
│ ├── mmdetection
│ │ ├── faster_rcnn_train.ipynb
│ │ ├── faster_rcnn_inference.ipynb
│ │ └── etc
│ └── requirements.txt
├── dataset
│ ├── test
│ ├── test.json
│ ├── train
│ └── train.json
└── sample_submission
├── faster_rcnn_mmdetection_submission.csv
├── faster_rcnn_torchvision_submission.csv
├── submission_ensemble.csv
└── train_sample.csv
- 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)
- Cascade RCNN
- YOLOv11
- Co-DINO
- RT-DETR
- Swin Transformer
- Resnet
| 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 |