Facilitating ADAS Validation Scale with Automated annotation


Implementing a new ADAS annotation help in ADAS system onto the road is an expensive and complicated undertaking. Testing an autonomous driving feature is a huge undertaking. It needs huge quantities of data from the perception system to be labeled with top quality. One of the largest limitations on time and budget is the manual portion in the process of labeling. Learn the way one auto Tier 1 supplier set a record speed when it came to creating image annotations the development and validation of a PoC for a computer vision software.

The Ambitious ADAS Development Timelines driven by OEM Customers

It was clear that the Proof of Concept project deadlines were determined by the development team's desire to show an unprecedented time for development and validation in their auto OEM clients. Only way for them to get the required throughput was to bring an annotation vendor to the project that could provide extremely automatized tool chain systems.

Key Validation Projects Figures

Annotation service that is fully managed

Annotation type 2D bounding boxes 7+ classes and attributes

Annotation volume: 27,000 km , with around 20 million objects

Annotation timeline The delivery time is 12 weeks.

Quality of annotation target The quality goal is 98%.

Delivery: 24-hour operation, including daily deliveries in batches

Annotation Supplier Challenge Annotation Supplier Challenge


There were a variety of requirements, including the automotive experience labels, the base of suppliers for labelers and tools that allow for a good collaboration between the supplier of data annotation as well as the in-house QA team.

Let's look at it in a different the context of. Based on the new average time to annotate The project team will require more than 22 weeks using the same labeling team size to finish an annotation for 23.4 million objects at the original speed of automation.

In this sequence you can observe the quality of object detection and accuracy of annotations produced completely automatically. There was no manual rework involved.

Why Automotive Experience is Important

The years of experience in large-scale annotation projects for autonomous driving has proven to be as crucial in how successful the venture is as machine learning expertise.

A common language and experience with common ADAS data collection or ADAS problems helped save time and avoided time lengthy clarifications.

UAI has experience in project management

The project was set up in record time

The operation was open 24/7 and included daily deliveries

Teams of 4 labels partners across the project.

Maintaining a high level of customer service regarding transparency in reporting as well as flexibility throughout the 12 weeks.

A network of labeling experts proved to be extremely useful in 2021. UAI was able to quickly create teams in locations that were not directly affected with Corvid. 

The Future of Autonomous Driving is Open to New Opportunities Driving Validation


The team that developed the Tier 1 has presented the Proof of concept result to its major OEM customers and received positive feedback in relation to performance and usability. This collaboration understand.ai has demonstrated the viability of projects where thousands kilometers must be driven to show that the autonomous vehicle's perception AI is operating accurately.

ADAS with GTS

To make this a reality, mechanization devices referenced prior in this blog can assist with accomplishing explanation at scale. Alongside this, you want a group that is sufficiently capable to empower information explanation at a huge scope. Are you considering outsourcing image dataset tasks? Global Technology Solutions is the right place to go for all your AI data gathering and annotation needs for your ML or AI models. We offer many quality dataset options, including Image Data Collection, Video Data Collection, Speech Data collection, and Text Data Collection.


Comments

Popular posts from this blog

Unlocking the Power of AI: Demystifying the Importance of Training Datasets

The Sound of Data: Unlocking Insights with Audio Datasets

What are the different types of AI datasets and how can they help develop the AI Models?