本篇文章希望可以總結一下常見的目标檢測基礎知識和煉丹技巧,以免在實際工作遇到問題的時候沒有辦法分析和解決
本篇部落格目錄
- 目标檢測:
- 常見trick:
-
- 1.必讀paper:Bag of Freebies for Training Object Detection Neural Networks
- 2. Loss:
- 3. Post-processing:
- 4. yolo系列解讀:
- 方法架構:
目标檢測:
目标檢測的是一個比較複雜的視覺任務,應用範圍很廣,不同的訓練方式都會對結果有很大的影響。
常見trick:
1.必讀paper:Bag of Freebies for Training Object Detection Neural Networks
文論解析通路!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2. Loss:
Generalized Focal Loss系列; 各種IoU Loss (GIoU, DIoU, CIoU)
3. Post-processing:
Soft NMS; Fast NMS; Cluster NMS; Matrix NMS; Box Voting; Iterative Pred;
4. yolo系列解讀:
yolov3 - yolov5,yolov7
yolox
方法架構:
方法架構
史前文明:DPM; SSVM
Two-stage:R-CNN; SPP; Fast R-CNN; Faster R-CNN; FPN; Mask R-CNN; Dynamic R-CNN; PANet; HTC; Double Head; Libra R-CNN; Cascade R-CNN
One-stage:YOLO系列; SSD; RetinaNet; ATSS; RetinaFace
Anchor-Free:FCOS; ATSS
End-to-end(NMS-free):POTO; DeFCN; PSS