- 1⃣️What are the differences between
anchor
andproposal
? - 2⃣️What are the differences between anchor-based methods and anchor-free methods?
- 3⃣️Baseline: the performance of faster-RCNN on Open Brand dataset.
- 4⃣️Read 2 anchor-based methods on CoCo and transfer to Open Brand.
- 5⃣️Read 2 anchor-free methods on CoCo and transfer to Open Brand.
- 6⃣️Read 2 transformer-based methods on CoCo and transfer to Open Brand.
- 7⃣️Read 2 papers about small object detection.
- 8⃣️Read 2 papers about long-tail distribution data.
- 9⃣️YOLOv5
- 🔟mAP (what, how)
Date | Heng Jia | Chenchen Han | |
---|---|---|---|
2021-04-12 ➡️ 2021-04-16 | 1⃣️: note 2⃣️: note 🔟: note |
9⃣️: 1 mAP on eval set | |
2021-04-17 | Report | Report | |
2021-04-18 | Summary and plan | Summary and plan | |
2021-04-26 10:00 | 2 验证集 top 1 model 验证集 top 10% model |
2 | 实时评测,每小时更新排行榜,每天两次提交机会 |
2021-06-03 15:00 | ==合并队伍==,初赛结束,共计 4 x 40 = 160 次提交机会 |
对抗攻击
- 训练集:58w images and 130w annotations
- 初赛测试集:5w images
- Read the paper open brands dataset
- EDA with pandas
- prepare codes for training
- Train LodoDet3K with YOLOv5s-default (will be completed on 22 April)