Hello Luxonis team,
I am currently working on an application that requires sub-millimeter depth accuracy at short distances (20–60 cm) using an OAK-D Pro camera.
I recently came across your blog post about Neural Depth and was wondering whether it would be possible to integrate Neural Depth models with the OAK-D Pro pipeline. Running the model directly on the device is not a requirement for my use case; I would be happy to run the depth inference on an edge computer or a desktop GPU if needed.
In my application, accuracy is significantly more important than latency. Even a depth error of around 5 mm can lead to major downstream issues, so I am looking for the most accurate depth estimation approach available for close-range operation.
Could you share whether Neural Depth can be used in this setup, and whether you expect it to provide a meaningful accuracy improvement over the standard stereo depth pipeline at distances between 20 and 60 cm?
Thank you for your time and insights. 😃
