AVCC and MLCommons Reveal Industry Benchmark for Automotive Perception with Cognata Synthetic Dataset

Cognata, a leading provider of simulation software and synthetic data for autonomous vehicles, proudly announces its significant contribution to the groundbreaking MLPerf Automotive Benchmark Proof-of-Concept (POC) released by MLCommons and the Autonomous Vehicle Computing Consortium (AVCC). This release marks a crucial milestone in developing a comprehensive benchmark suite for vehicle AI systems.

The POC, developed by the Automotive Benchmark Task Force (ABTF) comprising representatives from industry leaders like Arm, Bosch, NVIDIA, and Qualcomm Technologies Inc., includes a state-of-the-art SSD-ResNet50 object detector model trained on a robust dataset provided by Cognata. This collaboration highlights Cognata’s pivotal role in providing the synthetic data and know-how essential for advancing AI in the automotive sector.

Key Contributions by Cognata:

1. Extensive Dataset: Cognata provided a comprehensive dataset containing 120,000 8-megapixel images. These synthesized street-level views represent real-world scenarios, including complex and dangerous situations, enabling the creation of highly reliable AI models for collision avoidance systems.
2. Synthetic Data Expertise: Leveraging its expertise in generating realistic synthetic data, Cognata ensured that the POC dataset accurately reflects the dynamic and diverse environments encountered by autonomous vehicles.

The MLPerf Automotive Benchmark POC focuses on camera-based object detection, a crucial capability for collision avoidance and autonomous driving systems. By establishing common reference points for AI performance, the benchmark aims to help automotive OEMs and Tier 1 suppliers select and design optimal solutions for their systems, ensuring robust and reliable performance.

Read the full press release: https://mlcommons.org/2024/06/automotive-benchmark-poc/