Stéphane Charette

I welcome your PR. I've never built a CMake project that way, so I'm not familiar with FetchContent.

Can you post the output of the following: - `darknet --version` - `nvidia-smi` - `nvcc --version` Have you tried rebooting? Same thing?

In regards to this: > for darknet --version > Darknet V4 "Slate" v4.0-52-g5150ae50 > Darknet is compiled to use the CPU. GPU is disabled. Note you did not have CUDA...

> -- Looking for a CUDA compiler > -- Looking for a CUDA compiler - NOTFOUND Normally, the cause of this is that `nvcc` is not installed or not on...

Please post the output of `darknet --version`. Then also answer these questions: - the width and height of your neural network - the base configuration you are using - the...

Thanks for the information. > the error happens before the training starts Can you add `-trace` to the flags when you train? For example: darknet detector -trace -map -dontshow train...

Normally, the CUDNN_STATUS_BAD_PARAM error is due to out-of-vram problems, or algorithm selection. This normally means you don't have enough vram on your GPU. But it normally doesn't present exactly the...

I tried 224x160, 224x224, 160x160, and 128x128. The LEGO Gears network trained fine in all these cases, with no error. Here is an example with 224x224: I'll attach my 160x160...

I'm sorry, but I don't have a Quadro M1000M I can use to test. As I wrote, I tested (full training) with 224x160, 224x224, 160x160, and 128x128. I did not...

As discussed in the Darknet/YOLO discord, you should be checking out the branch like this: ``` git checkout --track origin/v5.1 ```