Metadata-Version: 2.1
Name: dataset-format-benchmark
Version: 0.0.10
Summary: Image dataset format benchmark
Home-page: https://github.com/kamikaze/dataset-format-benchmark
Author: Oleg Korsak
Author-email: kamikaze.is.waiting.you@gmail.com
License: gpl-3
Project-URL: Documentation, https://github.com/kamikaze/dataset-format-benchmark/wiki
Platform: any
Classifier: Development Status :: 4 - Beta
Classifier: Programming Language :: Python
Requires-Python: >=3.11
Description-Content-Type: text/x-rst; charset=UTF-8
Provides-Extra: cuda
Provides-Extra: testing
License-File: LICENSE
License-File: AUTHORS.rst

dataset-format-benchmark
========================

This package runs different image format benchmarks for dataset ML tasks

Installation
------------

Make sure you have some system deps installed:

.. code:: bash

   sudo apt install pkg-config libhdf5-dev

.. code:: bash

   python3.11 -m venv venv --upgrade-deps
   source venv/bin/activate
   python -m pip install -U -r requirements_dev.txt

   # For running on Nvidia GPU:
   python -m pip install -U torch torchvision

   # For running on CPU:
   python -m pip install -U torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu

   # For some reason h5py fails to install Cython while it needs it
   python -m pip install -U Cython

   python setup.py develop

Running dataset format benchmark
--------------------------------

.. code:: bash

   python -m dataset_format_benchmark --data-root /path/to/datasets/
