Skip to main content

Image Annotation Video Annotation Computer Vision Automotive

Advancing Autonomous Driving Through Innovative Annotation

 

The Challenge

Our client, a long-standing partner in the automotive industry, sought to train their autonomous driving algorithms to accurately identify various objects and elements at traffic intersections. The objectives ranged from recognizing simple objects, such as crosswalks and traffic lights, to interpreting more complex elements, like lane accessibility relative to a vehicle. The project presented unique challenges, including objects barely visible at a distance, which caused difficulties due to overlapping and perspective distortions, as well as undefined annotation cases for elements like roundabouts and bicycle lanes. Additionally, maintaining accuracy across a large volume of complex annotations required robust guidelines and processes.

• • • •The Solution• • • •

The DataForce team leveraged its proprietary platform and introduced several innovative approaches to ensure efficiency and accuracy:

  1. Training and Onboarding:
    • Annotators underwent detailed training sessions via video calls, with demonstrations, examples, and clear guidelines.
    • Sessions were followed by a test to verify understanding before gaining access to production annotation tasks.
  2. Workflow Optimization:
    • Tasks were segmented into Simple Experts for straightforward objects, and Complex Experts for advanced annotations, ensuring efficiency and focus.
    • Images were grouped by context and assigned to consistent pairs of annotators, preserving accuracy.
    • Workflow restrictors prevented task overload, ensuring fair distribution of work and reducing idle time across the team.
  3. Scripting:
    • Scripts were developed and integrated in the data pipeline, generating color-coded images, streamlining annotation workflows and enhancing quality.
  4. Multi-Tiered Quality Assurance:
    • A three-stage quality review process was established, including first- and second-level reviews, final annotation reviews, and internal expert assessments. This ensured precision and high-quality data transitions between teams.
  5. Client Scope Alignment:
    • Collaboration with the client clarified ambiguous cases and refined guidelines for consistent application.

Results

DataForce met and exceeded the client’s stringent requirements:

  • 98% Accuracy: Ensured minimal rejections and high-quality data delivery.
  • Increased Volume: Delivered over double the expected volume of simple objects while achieving complex object targets within 21% of the originally projected image count.
  • Budget Efficiency: Streamlined processes allowed the team to meet these goals while staying within budget, prompting the client to expand object counts and adjust scope ratios.

Thanks to DataForce’s advanced annotations and efficient workflows, the client was able to enhance their algorithms and achieve faster development timelines, ultimately advancing their ability to navigate complex traffic scenarios.

DataForce has a global community of over 1,000,000 members from around the globe and linguistic experts in over 250 languages. DataForce is its own platform but can also use client or third-party tools. This way, your data is always under control.

Request a consultation.