Benchmark IO speed of different backends

[Mr-Milk]

Goal

Benchmark the image reading speed of different backends using a PyTorch Dataset that loads image tiles.

Description

• The dataset will read and load tiles from images using different image I/O backends (e.g., OpenSlide, PIL, tifffile, zarr, etc.).
• Input images can be in various formats.
• For testing the OME-Zarr backend, input images will be converted to OME-Zarr format beforehand.

Tasks

• Define a standard tile sampling protocol for consistent benchmarking.
• Convert a representative set of test images into OME-Zarr format.
• Measure and compare tile reading time per backend.