
Pre-trained state-of-the-art real-time Vision Transformer model RT-DETRv2 (S, M, L and X). Options to either fine-tune existing model or load your own pre-trained model (additional fees may apply).

Can handle up to 4G pixels/second/channel. The number of channels varies depending on the selected model size and target FPGA board. Get instant feedback on the maximum pixel rate supported when selecting model and target FPGA board.

A C-based API to simplify interfacing the Machine Vision inference to an onboard RTOS. In a bare-metal implementation (no RTOS or onboard CPU) the user can manage the inference directly from the host PC.

A Python package to connect and manage the board from a host PC. The Python package includes a simple API to start/stop the inference as well as monitor the detection status and retrieve inference results.

Vendor and camera model agnostic interface that supports a configurable 8-bit or 16-bit raw pixels format (requires a camera that supports raw pixel interface).