I am reaching out to seek clarification regarding the implementation of the SpatialDetectionNetwork (specifically YoloSpatialDetectionNetwork) in the DepthAI ecosystem. In my current setup, I am using a DepthAI pipeline with mono cameras and stereo depth to detect objects and calculate their 3D coordinates, similar to examples provided in your documentation. However, I am encountering challenges in understanding how the SpatialDetectionNetwork processes bounding box data and depth information internally to produce accurate x, y, z coordinates for detected objects .
Could you please clarify whether the SpatialDetectionNetwork implementation, particularly the decoding and spatial calculation logic, is handled in firmware on the OAK device or if it is open-source and accessible in the DepthAI Python SDK or other repositories? If the implementation is part of the firmware, I would like to kindly request access to relevant documentation, code, or resources that would allow me to better understand and potentially customize the spatial calculations for my cotton picker project.