0

I am current training my custom dataset in a python environment with yolov5. A lot of my images I am targeting in my dataset are small therefore would like to train with - - 1280img. However when I try to run the code:

Pyhton train.py - - 1280img - - batch 32 - - epochs 100 - - data test123.yaml - - weights yolov5.pt

I get an error message>

RuntimeError: Unable to find a valid cuDNN algorithm to run convolution.

When I lower img down to 640 all is ok. But why can I not train in 1280img as if I am to target small objects and get the best results then its to train at this input. I assume its to do with maybe my GPU but I have a RTX 3080ti.

I appreciate your help

Thanks

Marze lells
  • 11
  • 3
  • 1
    Does this answer your question? [Unable to find a valid cuDNN algorithm to run convolution](https://stackoverflow.com/questions/61467751/unable-to-find-a-valid-cudnn-algorithm-to-run-convolution) – bro Jun 16 '22 at 02:32
  • 1
    Hi yes I reduced the batch size and it works now..... I did do this to begin with but from 64 to 32. Now its 16 from 32 and all is working. Thanks – Marze lells Jun 16 '22 at 02:59

0 Answers0