clarification on difference between visionworks and redtail?

[ghost]

Hi,

Both redtail and visionworks provide stereo camera depth detection. Just curious, why does NVidia have 2 repositories for this? Any general recommendations for picking one over the other? I get the feeling redtail is for video whereas visionworks is for static images?

[Alexey-Kamenev]

redtail's NVStereoNet was mostly experimental/research project to demonstrate the benefits of NVIDIA's Jetson platform for deep learning applications. Our original goal was to integrate the model with the rest of redtail code and make it a part of perception pipeline, but we did not have time to do that.

VisionWorks is a toolkit which contains a lot of different CV algorithms, including depth estimation from stereo. AFAIR, VisionWorks depth estimation uses SGM (i.e. not DL). VW should be easier to integrate into your code as that's the main purpose of the toolkit - be reusable.