HongChuang/ultralytics-ascend
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 3

Add missing HTML image alt tags (#6611)

Browse source 
Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com>
This commit is contained in:
Glenn Jocher 2023-11-27 17:46:29 +01:00 • committed by GitHub
parent 4096b261fc
commit 42bcf8c47f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
39 changed files with 267 additions and 245 deletions
Download patch file Download diff file Expand all files Collapse all files

12
docs/en/yolov5/tutorials/architecture_description.md
View file

@ -165,7 +165,7 @@ The YOLOv5 architecture makes some important changes to the box prediction strat
![b_w](https://latex.codecogs.com/svg.image?b_w=p_w\cdot&space;e^{t_w})
![b_h](https://latex.codecogs.com/svg.image?b_h=p_h\cdot&space;e^{t_h})
<img src="https://user-images.githubusercontent.com/31005897/158508027-8bf63c28-8290-467b-8a3e-4ad09235001a.png#pic_center" width=40%>
<img src="https://user-images.githubusercontent.com/31005897/158508027-8bf63c28-8290-467b-8a3e-4ad09235001a.png#pic_center" width=40% alt="YOLOv5 grid computation">
However, in YOLOv5, the formula for predicting the box coordinates has been updated to reduce grid sensitivity and prevent the model from predicting unbounded box dimensions.
@ -178,11 +178,11 @@ The revised formulas for calculating the predicted bounding box are as follows:
Compare the center point offset before and after scaling. The center point offset range is adjusted from (0, 1) to (-0.5, 1.5). Therefore, offset can easily get 0 or 1.
<img src="https://user-images.githubusercontent.com/31005897/158508052-c24bc5e8-05c1-4154-ac97-2e1ec71f582e.png#pic_center" width=40%>
<img src="https://user-images.githubusercontent.com/31005897/158508052-c24bc5e8-05c1-4154-ac97-2e1ec71f582e.png#pic_center" width=40% alt="YOLOv5 grid scaling">
Compare the height and width scaling ratio(relative to anchor) before and after adjustment. The original yolo/darknet box equations have a serious flaw. Width and Height are completely unbounded as they are simply out=exp(in), which is dangerous, as it can lead to runaway gradients, instabilities, NaN losses and ultimately a complete loss of training. [refer this issue](https://github.com/ultralytics/yolov5/issues/471#issuecomment-662009779)
<img src="https://user-images.githubusercontent.com/31005897/158508089-5ac0c7a3-6358-44b7-863e-a6e45babb842.png#pic_center" width=40%>
<img src="https://user-images.githubusercontent.com/31005897/158508089-5ac0c7a3-6358-44b7-863e-a6e45babb842.png#pic_center" width=40% alt="YOLOv5 unbounded scaling">
### 4.4 Build Targets
@ -204,15 +204,15 @@ This process follows these steps:
![match](https://latex.codecogs.com/svg.image?r^{max}<{\rm&space;anchor_t})
<img src="https://user-images.githubusercontent.com/31005897/158508119-fbb2e483-7b8c-4975-8e1f-f510d367f8ff.png#pic_center" width=70%>
<img src="https://user-images.githubusercontent.com/31005897/158508119-fbb2e483-7b8c-4975-8e1f-f510d367f8ff.png#pic_center" width=70% alt="YOLOv5 IoU computation">
- If the calculated ratio is within the threshold, match the ground truth box with the corresponding anchor.
<img src="https://user-images.githubusercontent.com/31005897/158508771-b6e7cab4-8de6-47f9-9abf-cdf14c275dfe.png#pic_center" width=70%>
<img src="https://user-images.githubusercontent.com/31005897/158508771-b6e7cab4-8de6-47f9-9abf-cdf14c275dfe.png#pic_center" width=70% alt="YOLOv5 grid overlap">
- Assign the matched anchor to the appropriate cells, keeping in mind that due to the revised center point offset, a ground truth box can be assigned to more than one anchor. Because the center point offset range is adjusted from (0, 1) to (-0.5, 1.5). GT Box can be assigned to more anchors.
<img src="https://user-images.githubusercontent.com/31005897/158508139-9db4e8c2-cf96-47e0-bc80-35d11512f296.png#pic_center" width=70%>
<img src="https://user-images.githubusercontent.com/31005897/158508139-9db4e8c2-cf96-47e0-bc80-35d11512f296.png#pic_center" width=70% alt="YOLOv5 anchor selection">
This way, the build targets process ensures that each ground truth object is properly assigned and matched during the training process, allowing YOLOv5 to learn the task of object detection more effectively.