Baidu data crowdsourcing platform

Scheme introduction

01 Data acquisition

The largest and most professional collection service team in China.

A large-scale integrated collection vehicle to provide cross city collection operations;
It has multiple sensor combinations such as Lidar, camera and radar;
Support customized vehicle and sensor modification;

02 Data annotation

The strongest data annotation capability in China, covering more than 10 different types of 2D/3D annotation tasks, accumulatively annotates more than 100 million frames of 2D/3D data in the field of automatic driving, with an accuracy rate of more than 99%.

Covers more than 10 different types of 2D/3D annotation tasks;
The largest labeling base in China, providing sufficient labeling manpower;
Industry leading intelligent auxiliary annotation technology and automatic quality inspection algorithm;

03 Privatized data annotation platform

Deployed in the customer's local area, the customer organizes employees or outsourcing personnel to annotate data on the enterprise intranet.

Provide comprehensive and powerful annotation tools to support function customization and docking with various systems;
Flexible and configurable project management process;
Hierarchical organization and personnel management;

04 Data management platform

Based on Baidu's huge experience in automatic driving data set management and training, it provides a data management platform to achieve efficient data storage, management and use.

Realize effective data management and sharing;
Realize model pipeline training operation;
Provide data cleaning services to eliminate invalid data;

05 Data training platform

The integrated training platform connects data, models and resources, and provides professional model evaluation services to help mine model defects.

Assembly line operation mode, simplifying model training process
Compatible with Apollo, Paddle, Caffe and other model frameworks
Tens of thousands of Cornercase mining model defects
Millions of kilometers of road test mileage, nearly 1000 scene categories, covering all aspects

Scheme advantages

Rich experience in data processing

In the field of automatic driving, it has collected tens of thousands of kilometers of data and labeled hundreds of millions of frames of data, and has rich experience in data acquisition and processing and large-scale data management training.

One stop data solution

It provides the most comprehensive intelligent driving data solution in the industry, covering all processes of data acquisition, labeling, storage, management and training, and realizes one-stop data management.

Leading data processing capability

Relying on Baidu's experience in automatic driving dataset processing, it provides data type processing services covering 2D, 3D, multi-sensor fusion and other data types, and has the industry-leading data acquisition capability.

Scientific data management platform

Based on Baidu's experience in data processing training, it launched three data platforms based on data annotation, management and training to achieve efficient data management and training.

Cooperation cases

Automatic driving solution provider
Automatic driving third-party evaluation mechanism
Automobile manufacturers

Automatic driving solution provider

3D obstacle framing

An autopilot solution provider provides the industry with autopilot technology based on laser radar. In the past, it used to label 3D point cloud data through a team of 100 people. However, due to the unstable data volume, which leads to high personnel maintenance costs, and the heavy self tagging mode, it is difficult to control the quality and productivity, so we seek cooperation with Baidu Intelligent Cloud.

Solution

Baidu Intelligent Cloud customized a special human resource plan for the labeling base for the customer according to the customer's half year data volume plan and cycle demand. The project manager led five team leaders and 80 experienced human resources for 3D point cloud labeling to serve the customer.

Programme results

The enterprise management work has been greatly reduced, the marking cost has decreased by 20% compared with the self built model, and the output and quality have been improved.

A third-party evaluation mechanism for automatic driving

2D semantic segmentation

A third-party evaluation organization for automatic driving provides technical evaluation services for the qualification of road test license plate examination for the governments of major cities. Since the data itself belongs to "examination questions", the requirements for the data confidentiality ability of the labeling company are very high. At the same time, the organization also provides data training for individual developers, and needs to develop a tagging platform. However, due to the heavy development workload, it seeks cooperation with Baidu Intelligent Cloud.

Solution

According to the requirements of customer confidentiality level, Baidu Smart Cloud provides customers with a privatization tagging platform+interface encryption service scheme. Help customers to build a marking platform and encrypt the interface at the same time, so as to realize the local data storage for customers, and open up the interface to be marked safely by Baidu Shanxi Base.

Programme results

The data security scheme was recognized by the customer, which solved the problem of insufficient capacity marked by the customer while ensuring the safety of customer data.
The deployment of privatization platform enables customers to not only have the ability to label tools, but also realize platform management.

An automobile manufacturer

2D/3D road data acquisition

An automobile manufacturer used to rely on Tier1 technology for L2 level auxiliary driving, but now the company has decided to build its own team to develop L4 level automatic driving. However, without Class A surveying and mapping qualification, it is impossible to legally collect 3D point cloud data on the road, and the manufacturer lacks annotation tools and capabilities, so it seeks to cooperate with Baidu Intelligent Cloud.

Solution

Baidu Smart Cloud has provided customers with an integrated solution for standard collection. It adopts the 64 line lidar+16 line lidar+FSAS collection scheme, and uses two self owned collection vehicles to collect a total of 2000 km of road data in Beijing and Shanghai.
The captured road video processing frames are extracted and transmitted to the customer for acceptance, and the acceptance will be marked online through the data through the security scheme of the privatization marking platform interface encryption. It also provides the human resource allocation scheme of Shanxi special human resources+resource platform, matching 1200 base human resources+1000 resource human resources with relevant experience for various types of 2D and 3D tasks of customers, combining the platform process and tools, to achieve the double guarantee of quantity and quality.

Programme results

3D box selection: 28W boxes are produced daily, with an accuracy rate of 99%; 2D segmentation: 4000 frames per day, with an accuracy rate of 98.4%;
2D box selection: 106W boxes produced daily, accuracy rate 99%

Integrated intelligent driving data solution

Data crowdsourcing service

Scheme architecture

01 Data acquisition

02 Data annotation

03 Privatized data annotation platform

04 Data management platform

05 Data training platform

Scheme advantages

Rich experience in data processing

One stop data solution

Leading data processing capability

Scientific data management platform

Cooperation cases

Automatic driving solution provider

Solution

Programme results

A third-party evaluation mechanism for automatic driving

Solution

Programme results

An automobile manufacturer

Solution

Programme results

Professional AI data helps the development of enterprise intelligence