Open JoeCool90 opened 5 years ago
I didn't test non-gpu training for a long time, since it is very slow. It will take whole life.
May be it is better to use OpenCL Darknet implementation to train on AMD GPU:
Thanks for the suggestions and sorry I couldn't help debug further. I've given the second link a go as it has a msvc project, but it has some buffer management issues which means it bugs out on my 580. I'll give the first one a try in linux.
I've tried training, first for a custom object, and then back to basics following the Pascal Voc training tutorial.
Every time, I'm getting a memory exception in visual studio, or seg fault on linux, which VS or gdb identify as coming from im2col_get_pixel.
im seems to be valid, but the row/height/channel index into an area that hasn't been allocated. I've had a bit of a look, and tried compiling with and without opencv support, just in case it was an image loading problem. I'm using the no-gpu option as I have an AMD card.
Thanks for any help you can give, Joe