fast image resize about kornia-rs

[edgarriba]

Explore this crate and benchmark: https://docs.rs/fast_image_resize/latest/fast_image_resize/index.html

DONE

[ternaus]

Darwin Vladimirs-MacBook-Pro.local 23.2.0 Darwin Kernel Version 23.2.0: Wed Nov 15 21:53:18 PST 2023; root:xnu-10002.61.3~2/RELEASE_ARM64_T6000 arm64

kornia_rs==0.1.2
numpy==1.26.4
opencv-python-headless==4.9.0.80
pillow==10.2.0

python py-kornia/benchmark/bench_resize.py
OpenCV: 0.04 ms
PIL: 0.20 ms
Kornia: 0.17 ms

I believe that the benchmark should run over many different images.

This setup could be a bit too extensive, but still....
https://github.com/ternaus/imread_benchmark/blob/main/imread_benchmark/benchmark.py

[edgarriba]

I think this benchmarks should be running in a Linux machine which I guess is the most typical scenario for training. Besides, why don’t you use timeit which run several iterations and remove any possible outlier plus other things for you. Also, the benchmark should really be done against a data loader routine and measure probably IPS (images per second)

[ternaus]

Just run your code from https://github.com/kornia/kornia-rs/blob/main/py-kornia/benchmark/bench_resize.py

Would be happy to check a version with an updated setup for a benchmark :)

[ternaus]

Do you have comparison with other libraries, somthing similar to:

https://github.com/albumentations-team/albucore/blob/main/benchmark/results/uint8_3.md

and if resize in kornia-rs is not broken, as it is in torchvision, tensorflow or OpenCV it would be ultra great.https://github.com/assafshocher/ResizeRight