odl-cli 0.1.1.post143

 

OpenDataLab website ^HOT     

 

English | 简体中文

📘Documentation |

Introduction

Data is the cornerstone of artificial intelligence. The efficiency of data acquisition, exchange, and application directly impacts the advances in technologies and applications. Over the long history of AI, a vast quantity of data sets have been developed and distributed. However, these datasets are defined in very different forms, which incurs significant overhead when it comes to exchange, integration, and utilization -- it is often the case that one needs to develop a new customized tool or script in order to incorporate a new dataset into a workflow.

To overcome such difficulties, we develop Data Set Description Language (DSDL).

Major features

The design of DSDL is driven by three goals, namely generic, portable, extensible. We refer to these three goals together as GPE.

• Generic

  This language aims to provide a unified representation standard for data in multiple fields of artificial intelligence, rather than being designed for a single field or task. It should be able to express data sets with different modalities and structures in a consistent format.

• Portable

  Write once, distribute everywhere. Dataset descriptions can be widely distributed and exchanged, and used in different environments without modification of the source files. The achievement of this goal is crucial for creating an open and thriving ecosystem. To this end, we need to carefully examine the details of the design, and remove unnecessary dependencies on specific assumptions about the underlying facilities or organizations.

• Extensible

  One should be able to extend the boundary of expression without modifying the core standard. For a programming language such as C++ or Python, its application boundaries can be significantly extended by libraries or packages, while the core language remains stable over a long period. Such libraries and packages form a rich ecosystem, making the language stay alive for a very long time.

Installation

Case a install it with pip

pip install dsdl

Case b install it from source

git clone https://github.com/opendatalab/dsdl.git
cd dsdl
python setup.py install

Get Started

Use dsdl parser to deserialize the Yaml file to Python code

dsdl parse --yaml demo/coco_demo.yaml

Modify the configuration & set the directory of media in dataset

Create a configuration file config.py with the following contents（for now dsdl only reading from aliyun oss or local is supported）：

local = dict(
    type="LocalFileReader",
    working_dir="local path of your media",
)

ali_oss = dict(
    type="AliOSSFileReader",
    access_key_secret="your secret key of aliyun oss",
    endpoint="your endpoint of aliyun oss",
    access_key_id="your access key of aliyun oss",
    bucket_name="your bucket name of aliyun oss",
    working_dir="the relative path of your media dir in the bucket")

In config.py, the configuration of how to read the media in a dataset is defined. One should specify the arguments depending on from where to read the media：

1. read from local： working_dir field in local should be specified (the directory of local media)
2. read from aliyun oss： all the field in ali_oss should be specified (including access_key_secret, endpoint, access_key_id, bucket_name, working_dir)

Visualize samples

dsdl view -y <yaml-name>.yaml -c <config.py> -l ali-oss -n 10 -r -v -f Label BBox Attributes

The description of each argument is shown below:

Citation

If you find this project useful in your research, please consider cite:

@misc{dsdl2022,
    title={{DSDL}: Data Set Description Language},
    author={DSDL Contributors},
    howpublished = {\url{https://github.com/opendatalab/dsdl}},
    year={2022}
}

License

DSDL is released under the Apache 2.0 license.

Acknowledgement

• Field & Model Design inspired by Django ORM and jsonmodels