DEPRECATION WARNING

This repo is currently not actively maintained. New features and formats within the Encord application might break the code here.

Please contact [email protected] if you have any questions. You can also see the Encord python SDK documentation to learn more about how to fetch data from the Encord platform.

---

The data engine for computer vision

This repository holds an implementation of a Dataset, which is compatible with the PyTorch framework. The Dataset will fetch information about data, geometries, and labels from the cord-client-python module and structure it into an easily accessible format.

Installation

$ python -m pip install git+https://github.com/encord-team/encord-dataset.git

Other requirements

The data loader uses ffmpeg to split videos into individual frames. So make sure that you have ffmpeg installed:

$ ffmpeg -version

Usage

from encord_dataset import EncordData

config_path = './example_config.ini'
data = EncordData(config_path, download=True)

img, attr = data[0]
print(attr)

Configuration

The Dataset depends on a project_id, api_key, and a cache_dir, which can all be specified in a <your-config>.ini file, as done in example_config.ini.