Skip to content

Commit

Permalink
目标检测常见问题
Browse files Browse the repository at this point in the history
  • Loading branch information
315386775 committed Nov 27, 2024
1 parent 0f937be commit 60263e4
Show file tree
Hide file tree
Showing 2 changed files with 9 additions and 1 deletion.
1 change: 1 addition & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -180,6 +180,7 @@
| [**23. 场景问题:通过设置更多的先验anchor能否改善小目标及非正常尺寸目标的性能,除计算速度外还存在什么问题**](VisionPerception/Reference.md) |
| [**24. 场景问题:目前检测常需要NMS非极大值阈值算法作为后处理,有没有什么可以避免NMS后处理的方案?**](VisionPerception/Reference.md) |
| [**25. 模型问题:如何理解DETR中的object query的概念,要为 cross attention 提供更好的位置先验该如何设计模型?**](VisionPerception/Reference.md) |
| [**26. 模型问题:YOLOV5和YOLOV8的Head输出通道数分别是多少,假设现在是2个类别的检测任务?**](VisionPerception/Reference.md) |

- 目标分割

Expand Down
9 changes: 8 additions & 1 deletion VisionPerception/Reference.md
Original file line number Diff line number Diff line change
Expand Up @@ -409,4 +409,11 @@ def process_output(output):
# 25. 模型问题:如何理解DETR中的object query的概念,要为 cross attention 提供更好的位置先验该如何设计模型?

- 位置部分则来自于 learnable queries;
- 引入anchor box作为query提供位置先验,可参考DAB-DETR;
- 引入anchor box作为query提供位置先验,可参考DAB-DETR;

# 26. 模型问题:YOLOV5和YOLOV8的Head输出通道数分别是多少,假设现在是2个类别的检测任务

```python
self.reg_max = 16 # DFL channels (ch[0] // 16 to scale 4/8/12/16/20 for n/s/m/l/x)
self.no = nc + self.reg_max * 4 # number of outputs per anchor
```

0 comments on commit 60263e4

Please sign in to comment.