CUDA ERROR OUT OF MEMORY

CUDA_ERROR_OUT_OF_MEMORY occurred in the process of following the example below.

Object Detection Using YOLO v4 Deep Learning - MATLAB & Simulink - MathWorks 한국

No changes have been made in the example.

The GPUs used are GeForce RTX 3060 VRAM 12GB and 32GB RAM.

I wonder if my equipment is insufficient or if I need to adjust the options.

[Walter Roberson]

I have found statements online that YOLO v4 takes a lot of memory. I would have thought your system was large enough, though.

Try reducing the minibatch size. A paper I found online said that for YOLO v4, the optimal minibatch size is 2 or 3, and beyond that you do not get any performance or useful accuracy gains.

[은석 최]

Thank you for your advice! I set the mini-batch size to 2, so it's resolved.

[Jan]

By the way: This is not tiwtter - no # before the tags. "error" is not useful as a tag.

Answer 1

[Nihal Reddy]

It looks like a memory issue. Please try resetting your device using the command-

(reset(gpuDevice)) 

then put a breakpoint before the call to “yolov4ObjectDetector” function and check the available memory.

If the error persists, then check the size of input data and change the mini batch size accordingly to the memory available.

You can also query or select a GPU device using “gpuDevice” function.