This project presents an FPGA architecture for the computation of visual attention based on the combination of a bottom-up saliency and a top-down task-dependent modulation streams. The target applications are ADAS (Advanced Driving Assistance Systems), video surveillance, or robotics.