search

sovit-123 / fasterrcnn-pytorch-training-pipeline

PyTorch Faster R-CNN Object Detection on Custom Dataset
MIT License
223 stars 75 forks source link

Selection of the number of --batch sizes for the training process #116

Closed auliadarnalisariii closed 10 months ago

auliadarnalisariii commented 11 months ago

How do I choose the number of --batch sizes for the number of epochs I use??

Please provide me the best solution for its use

sovit-123 commented 10 months ago

Generally, it is chosen based on trial basis. Run the training a few times and keep increasing the batch size starting from a small value each time. You should provide batch size such that is an even number and consumes the maximum memory of your GPU.

sovit-123 commented 10 months ago

Hi, I hope the issue is solved. I am closing the issue because of no activity for now. Please reopen if needed.