I have questions about OAK-D and OAK-D-Lite. Please check this out and let us know. ^^ 1. How accurate is the depth estimation compared to Intel Realsense D435? 2. Specification of AI / OpenCV-based solutions (in the case of object detection, mAP or dataset-based Reference based test result from an empirical point of view, etc.) 3. Is it a PoC product as a KickStarter or is it a formally released product? 4. In above three questions, If anything is different between OAK-D and OAK-D-lite, please point out what is different those hardware to select right hardware for our project. Thanks.

Hello @"RyanLee"#p2291 , great questions! 1. I would say that at perfect lighting conditions, the depth accuracy is about the same (below 2%). In-doors, D435 will have better depth accuracy though, as there is an IR laser dot projector, so you should compare D425 with [OAK-D-Pro](https://github.com/luxonis/depthai-hardware/issues/114), which also has IR dot projector. We have initial prototype of the OAK-D-Pro and are working on eye laser-safety certification atm. [Here is](https://docs.google.com/document/d/1C4_9C6ulL7cMPhZrXHU6SSjqcs05s_apK8NWIu6HkxI/edit) a comparison table between Realsense and our devices. 2. We are actually currently working on this. Here are the initial results. In short, it should be approximately the same as the AI results on Intel's NCS2, as the chip is the same (Myriad X). [URL=https://i.imgur.com/tsdyecn.png][IMG]https://i.imgur.com/tsdyecn.png[/IMG][/URL] [URL=https://i.imgur.com/RFIUfMn.png][IMG]https://i.imgur.com/RFIUfMnh.png[/IMG][/URL] 3. Both products (OAK-D and OAK-D-Lite) are in "release state", but OAk-D-Lite hasn't been mass-manufactured yet (will be in the following month, almost 10k OAK-D-Lites). 4. Differences are [described here](https://docs.luxonis.com/projects/hardware/en/latest/pages/DM9095.html#oak-d-vs-oak-d-lite). Thanks, Erik

OAK-D and OAK-D lite

RyanLee

I have questions about OAK-D and OAK-D-Lite. Please check this out and let us know. ^^

How accurate is the depth estimation compared to Intel Realsense D435?
Specification of AI / OpenCV-based solutions (in the case of object detection, mAP or dataset-based Reference based test result from an empirical point of view, etc.)
Is it a PoC product as a KickStarter or is it a formally released product?
In above three questions, If anything is different between OAK-D and OAK-D-lite, please point out what is different those hardware to select right hardware for our project.

Thanks.

erik

Hello RyanLee , great questions!

I would say that at perfect lighting conditions, the depth accuracy is about the same (below 2%). In-doors, D435 will have better depth accuracy though, as there is an IR laser dot projector, so you should compare D425 with OAK-D-Pro, which also has IR dot projector. We have initial prototype of the OAK-D-Pro and are working on eye laser-safety certification atm. Here is a comparison table between Realsense and our devices.
We are actually currently working on this. Here are the initial results. In short, it should be approximately the same as the AI results on Intel's NCS2, as the chip is the same (Myriad X).
Both products (OAK-D and OAK-D-Lite) are in "release state", but OAk-D-Lite hasn't been mass-manufactured yet (will be in the following month, almost 10k OAK-D-Lites).
Differences are described here.

Thanks, Erik

RyanLee

Thanks Erik,

Now somehow i got a sense for this except one part. Could you explain two pictures in the second answer?
Thanks again.

Best regards,

Matija

Hi RyanLee ,

What @erik linked are just the beginnings and currently it is only the FPS estimations, as well as the number of parameters and FLOPs. We are in the process of migrating the models from experiments (+ we are constantly looking for new models) to the model zoo. With that in mind, we also opened a new issue that will do exactly that. Evaluate the models on known data sets with execution on OAK, and then compare the performance. This will also help us discover potential bugs and improve the performance / conversion process. Help from the community is appreciated here, and I'll be writing the evaluation script in the next few days.

Right now you can find some results for mAP and other standard evaluation metrics for models from OpenVINO model zoo (e.g., yolo-v3-tiny-tf), though I am not sure if they provide the results of inference for OpenVINO IR or original results provided for models in Pytorch, TF, ... (so, before conversion). Nor am I sure whether they use FP32 or FP16.

That said, a slight drop is expected on OAK devices, as FP16 is used.

erik

Asking Matija, our ML engineer, to answer your question.

RyanLee

Matija Oh, thanks Matija. I am checking about your feedback. if there is any more question i will come back to you. 🙂