Visit Baidu AI

Declaration of Data Open and Sharing

I.  Apollo Manifesto

Open Capability--Shared Resources--Accelerated Innovation--Sustained Mutual Benefit

II.  Data Sharing

1. Principle of fair data

  • •  In terms of the data open scale, it follows the principle of “the more you contributions, the greater you achievements”.
  • •  The contributions of partners are calculated based on data mileage (km) as the unit.
  • •  The achievements refer to the data, services and algorithm models that partners can obtain in the Apollo Open Data Platform (hereinafter referred to as this “Platform”).

2. Principle of data storage

•  Data collected in China shall, and shall only, be stored on servers in China; data collected in other countries and regions shall be subject to data storage restrictions specified by laws of the relevant countries.

3. Provision of data

  • •  All data must be uploaded in the format which meets the Apollo data specifications.

4. Initial data and subsequent data

  • •  As the initiator of this Platform, Baidu provides the initial data for this Platform. The data will be open to all partners of this Platform.
  • •  After the initial data, this Platform will develop gradually relying on the contributions of all this Platform partners. Each partner shall follow the principle of fair data.

5. Data processing

  • •  This Platform has the right to process the data uploaded and marked as public by partners. The processed data will be used as this Platform’s public data.

6. Data privacy

  • •  Any partner can view their own data and set the privacy properties of the data as: private or public.
  • •  Private: it indicates that the data can only be viewed and used by the partner itself.
  • •  Public: it indicates that the data is licensed to this Platform. Platform will disclose the derived data and resources based on the principle of data sharing of this Platform.
  • •  The data uploaded by partners are considered to be private by default. The data after the secondary processing by Apollo based on the data uploaded and marked as public by partner (such as annotation or scene extraction) should be used as the public data of this Platform.
  • •  Blacklist: if partners do not want their data to be accessed by any sensitive partner, they can make the relevant settings.
  • •  Special data: which is considered to be the private data of partners by default. The owner of the data can specify the data open to specific partners of this Platform.

7. Data access

  • •  Partners can use this Platform to view the data.
  • •  Partners can train their own models on this Platform, and access and save the data via this Platform API during the training.
  • •  The calculation process data “beyond this Declaration” generated during the model training are owned by the partners themselves.

8. Calculation results

  • •  The “calculated results” (e.g., calculation process data, calculation results, logs, data models, etc.) obtained by the partner through this Platform and data calculations are owned by the partner, and can be downloaded by the partner itself freely. Other partners cannot view or access the results.

III.  Data Scope and Open Method

1. Data Scope

1) Original data

It refers to the data generated during the running of the vehicle, including sensor data, driving behavior data, positioning data, etc.

2) Annotation data

It refers to the manual or mechanical annotations based on the original data, such as 2D / 3D vehicle annotations, pedestrian annotations, lane annotations, special driving scene data, etc.

3) Automatic Driving Scene (ADS)

  • •  This Platform will provide simulation simulator, as well as various simulation scenes that can be run on the simulator.
  • •  Baidu Smart Car will serve as a partner of this Platform to provide the first batch of common classic scenes for this Platform.
  • •  Each partner of this Platform can also use the simulator to edit the scenes, or share the scene data of the automatic driving to this Platform.

4) Special data

For example: engine adjustment table (e.g. how much acceleration can be got based on the use of the accelerator), CAN bus protocol, etc.

2. Open method

  • •  The partners will be provided a small amount of each type of demo data to download locally for program debugging.
  • •  The annotation data and automatic driving scenes (ADS): partners can view the data online via this Platform, or deploy the algorithm program to this Platform to call the data via this Platform API.
  • •  The special data can only be viewed, and cannot be accessed by API or downloaded through this Platform.

IV.  Data Contributions

The development of this Platform relies on the joint efforts of all Platform partners. When visiting the public data provided by this Platform, partners are also welcomed to actively share their own data on this Platform by us. We will adhere to the principle of fair data to make sure that partners with greater contributions will obtain more data and services from this Platform.