• DepthAI-v2

  • Head-to-head comparison of OAK-D and OAK-D-Lite?

I would love to see a breakdown of all the differences between OAK-D and OAK-D-Lite. I mean something like this as you did before.

Moreover, since everything in engineering is a trade-off, I wonder the benefits of being smaller/lighter/cheaper come at what cost?

  • erik replied to this.

    Hello nalzok, the main reason for smaller/lighter (and somewhat cheaper) is that we used Myriad X VPU directly on the PCB, instead of connecting the SOM to the baseboard (like on OAK-D). Also, we made it smaller (and therefore lighter, less enclosure) by developing (with ArduCam) custom camera sensor connectors that are shorter and take up less space. Also another reason for the better price point is the ability to get the bulk buy discount, as we will be manufacturing a lot more of these devices.
    Tradeoffs:

    • Cameras aren't as high-quality. Mono cameras have smaller resolution (640x480 instead of 1280x800).
    • There (most likely) won't be IMU on the OAK-D-Lite. These are really hard to get these days and prices went waay up.
    • (Not that relevant) No power jack, as most users actually just used USB-C for power delivery as well.

      Sounds like a well deliberated great hardware refresh. According to the (temporary?) specs out there it seems that the upcoming lite model will also sport a slightly wider FOV and a somewhat closer recommended minimum distance than the OAK-D, if my record keeping is correct.

      • erik replied to this.

        Hello matan, both of your observations are correct, so your record-keeping is correct🙂

            Thanks erik , that's good to know. Specifically, it seems that the minimum recommended distance will be down from 19.6 cm to just 8 cm if I'm reading it right, but I would naively assume that will primarily be a smaller minimal focus distance for the lenses, and not necessarily an indication of high accuracy depth detection at such close ranges approaching that new minimal recommended distance that the lite model will probably have.

            • erik replied to this.

                You are correct matan. In theory, with downscaling to 240P (instead 480P) and using extended-disparity, the theoretical minimum distance would be ~ 8cm. I believe that at such short distances, the (left/right) mono frame overlapping is minimal (like 20%) so most of the depth map would be just blank - with some depth estimation on the (left) side of the frame (since disparity is calculated from right mono camera). I hope my explanation makes sense🙂