• DepthAI-v2
  • Is the disparity calculation open source?

I had a look, but have not found the disparity calculation code, or other parts of the firmware for the OAK-D camera on github. Is it open source?

  • erik replied to this.

    Hello philip , the firmware itself isn't open-source, as anything related to it (build tools etc.) are considered Intel's IP - same is for any PCB that has Myriad X on them (eg. OAK SOM). The disparity calculation can also be done on host but if you are just curious about algorithms, it's described here.
    Thanks, Erik

    Thanks, that's helpful. We are considering how we could use OAK cameras. One of the primary uses would be as an alternative to a D435 for obstacle avoidance on a mobile robot. Just trying to understand what avenues are available if we need to improve aspects of depth estimation.

    • erik replied to this.

      Hello philip , I believe we expose all (or at least majority) of settings of the stereo depth HW block, so it is very customizable. You can see all available settings here (under reference). We are also currently working on on-device filters (temporal, threshold, speckle.. PR here), so depth perception can be better/less noisy. Thoughts?
      Thanks, Erik

      Hi Erik. Thanks for the tips. I played with the existing parameters, and will give the in progress filters a try at some point. The quality of point clouds we get out of a D435 is barely adequate for our application with a good amount of filtering. Looking at the output of the Oak-D, I am skeptical that we will get adequate point clouds out of it. One obvious issue is the lack of a projector and consequent lack of depth estimate, or poor depth estimate for parts of the scene with little texture. I know you guys are working on a product with a projector, or we could add a projector ourselves. However, even ignoring the lack of texture related issues, I've found it hard to balance having decent depth estimate fill rate while not having significant spurious depth estimates when playing with the existing parameters. I'll keep evaluating though, and we may also find a use that is more in line with the spatial AI stuff that you guys appear to focus on.

      • erik replied to this.

        Hello philip , could you share some comparison footage/tests? We are currently focusing a lot on depth quality, so any insight would be helpful. From what we have tested, it's on par with passive stereo depth of D435.
        Thanks, Erik

        Hi Erik, sure no problem. I'm working remotely and won't have a D435 to compare side by side until mid January. I have an Oak-D and am comparing it to my expectation of how a D435 would work for the moment. I'll send a more scientific comparison some time in January.

        Thanks,
        Philip

        • erik replied to this.

          Hello philip , that sounds great, thanks! We are also working on RealSense comparison docs (PR) that will soon get merged, in case you are interested.
          Thanks again, Erik