BossSensor
BossSensor copied to clipboard
detecting problem in finally
I have accomplish the model's train and evaluate,the evaluate result is 93%,but when I predict face on my camera , the process will think of all the faces as me . I should how to the problem。
I stongely agree with you. And I have a simialr problem. So, I think I need to modify the CNN model.....
I find out that if you increase epoch to about 600, the accuracy will increases largely. Also, I am agree that adding some more layers to the CNN model will help. I have made a project called FaceLock inspired by this project. And the accuracy is acceptable. Check it out on https://github.com/Donny-Hikari/FaceLock
I recently use Donny's model, which works more effective. 💯
After training for about 40 times,the accuracy is stable at 0.88.