Albumentations vs Kornia Benchmarks
This route is reserved for benchmark results generated during the website build.
The benchmark implementation source is public:
- Benchmark transforms directory
- Kornia implementation
- Kornia multichannel implementation
- Kornia video implementation
- Shared transform specs
Methodology Notes
The generated benchmark page should make the execution model explicit. Kornia benchmarks are tensor benchmarks; Albumentations benchmarks are usually NumPy array pipeline benchmarks.
The generated output should separate:
- image transforms
- multichannel image transforms
- video or batched transform paths
- pipeline benchmarks
- CPU versus GPU execution when GPU numbers are included
For Kornia, batch size and device placement can change the conclusion. A GPU transform that is fast in isolation can still hurt training if it competes with the model for device time.