rabbitAI provides device-specific, high-precision ground truth solutions for computer vision engineers developing 3D perception and recognition algorithms. This specialized training data accelerates algorithm learning, enabling efficient specialization on target hardware. The outcome is improved inference quality, reduced hardware costs, and faster development cycles for embedded AI systems.
Funding
Funding not disclosed
Founders
Product
Problem
Developing reliable 3D perception, action, and gesture recognition on low-end hardware for XR or in-cabin sensing solutions is challenging. Meeting functional safety requirements demands rigorous validation, but efficiently collecting the necessary validation data is difficult. Existing methods struggle to capture ultra-realistic edge cases and critical scenarios, hindering the accuracy and reliability of AI algorithms.
Solution
The company provides device-specific ground truth solutions that capture complex and dynamic real-world scenarios with high precision. Their technology accelerates the learning process of AI algorithms, allowing them to specialize efficiently on specific hardware. By extracting 3D models from real-world image data with millimeter-level resolution, they derive automated, pixel-perfect ground truth for images from real camera devices. This approach enables more room for optimization, reduces hardware costs, and shortens development cycles. The solution also includes the ability to augment existing real-world data with synthetic components, maximizing the value of acquired data and enabling virtual testing of new devices.
Target Audience
The primary customers are computer vision engineers, OEMs, and Tier 1 suppliers developing in-cabin sensing and VR/AR applications, particularly those working with low-end hardware and facing stringent functional safety requirements.
Features
- Mobile and flexible capture solution for smooth operations and fast integration.
- High scenario variance and throughput.
- Dense depth and body poses at 1mm precision.
- Key points, object segmentation, and other modalities, all automated.
- Robust ground truth against lighting, mechanical shocks, and other environmental factors.
- Dynamic scenarios captured at 60fps, agnostic to indoor and outdoor environments.
- Virtual testing of new devices with different extrinsics and intrinsics using close-to-real camera simulation on recorded image data.
- Edge-case augmentation of real data, such as tattoos or sunglasses on human faces, to increase data variance.