Train faster-rcnn with multiple images per iteration

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hi, I find training super slow even with two GPUs, and for a small model like ResNet18. My images are 1280x720. But PyTorch trains much faster with the same image size as I use multiple images per each iteration. Does tlt-train allow this? Thanks

2

Hi chandrachud,
See https://docs.nvidia.com/metropolis/TLT/tlt-release-notes/index.html,
FasterRCNN currently supports only single GPU training with a batch-size of 1.

3

Thanks for the reply.
I see on this link How to Build and Deploy Accurate Deep Learning Models for Intelligent Image and Video Analytics | by NVIDIA AI | DataSeries | Medium that detectnet supports multi-gpus.
It’s not clear though if batch sizes > 1 are supported and how to set this parameter in the spec file?
Thanks

p.s. By the way, it would be nice to have a search feature on this forum!

4

Hi chandrachud,
Yes，detectne_v2 supports multi batches. You can set “batch size” in the training spec.
More, the forum has searching function.You can click the button on the top right corner, and type the key word, select which specific forum to search.