Detection works fine on CPU but generates random boxes (wrong or no detection) on GPU

Question

Detection works fine on CPU but generates random boxes (wrong or no detection) on GPU

barkhaneum opened this issue 3 years ago · 6 comments

barkhaneum commented 3 years ago

Hello, I'm testing in RTX 3080ti environment. Is there anyone who succeeded?

Answer 1 · 2022-01-15T08:07:25.000Z

same with me. how to solve it?

Answer 2 · 2022-01-15T10:44:02.000Z

What is the error that is being thrown?

Answer 3 · 2022-01-17T01:00:11.000Z

What is the error that is being thrown?

the GPU works well but no bounding box detect to the object.

Answer 4 · 2022-01-17T02:05:47.000Z

I looked for various problems to solve the problem, and I think the CUDA version and Tensorflow version used in 3080ti are incompatible. As I lowered the Tensorflow version, I think it's a problem related to numpy. So I used the method of training on darknet with 3080TI and detecting with 2080super. To solve this problem, we need to fix the source again with the latest version of Tensorflow.

Answer 5 · 2022-01-21T16:17:54.000Z

look at #77 to solve the problem

Answer 6 · 2022-03-08T05:17:37.000Z

look at #77 to solve the problem
Thank you it works perfectly for me :)