[PYTHON] Detect a stopped sign.

at first

Until the last time, I made a model that recognizes Nogizaka's face, but this time I made a program to detect rare signs. I will make it. Ultimately, I will aim for a system like Edge AI.

About environment construction

I wrote Nogizaka's face recognition program until the last time, and I will use exactly the same one that I used there, so I hope you can see the environment with reference to that.

About annotations

This is the same as last time, so please.

About learning

We will train using 70 images. The annotated image looks like this.

epochs 100 times

result

Conclusion

I feel like this is the case with about 70 images, but I can't actually use it. However, I would like to improve the model after successfully incorporating this into the system, so I will postpone the improvement of the model. Next, I would like to write a visualization part that applies to other programs.

Postscript

I also tried to learn with epochs 1000. The result is here. I tried to verify with the same image this time as well. The result is here. You can see that it is repelling firmly, so this learning was successful. Next, I will modify yolov5 a little using this model.