performance/2019/05/02/pytorch-augmentation-with-libjpeg-turbo

[utterances-bot]

Squeezing image loading performance using libjpeg in Pytorch | Basic Machine Learning

Sometimes in your project you seek cheap thrills out of changing couple of lines of code. This is what happened when I came across the libjpeg-turbo project. The turbo version of libjpeg boasts to be 2-6x faster in performing image processing operations. Performance is what everyone wants, right?

https://www.basicml.com/performance/2019/05/02/pytorch-augmentation-with-libjpeg-turbo.html

[OlofHarrysson]

Thanks for the post. I'll remember this if I ever get my hands on a GPU good enough to bottleneck the CPU :D

Just because you are so fond of the data augmentation + reading I thought that you might be interested in a drop-in replacement for pillow that is sometimes faster -> pillow-simd.

Or by imgaug, a library that can do some really cool image augmentations!

[pbamotra]

Hi @OlofHarrysson

Thanks for the suggestions. I'm currently tweaking Nvidia DALI to make image augmentation faster. Basically, just doing everything on GPU than on GPU. I used pillow-simd, but as you said, the performance improvement happens only sometimes.

I'll checkout imgaug, I starred it a while back, didn't check it out thought.

Also, thanks for reading my blog. =)