NuitrackAI info


#1

Hello,

I would like to know information about the new Nuitrack AI.

  • When a beta or release will be available ?
  • Do you provide possibility to update learning data with our data ?
  • The output will be joint 3D positions and orientations ?
    Thank you a lot.

Regard,

PP


#2
  • When a beta or release will be available ?
    in following 1-2 months
  • Do you provide possibility to update learning data with our data ?
    yes, at this stage we can fine tune for your case if you send us *.oni files of your case to email-nuitrack(at)3divi.com
  • The output will be joint 3D positions and orientations ?
    yes

#3

To provide more details:

This will be closed/canary testing of beta version, to participate please send request through the feedback form.

Strictly speaking, no. At least not in the process of regular usage of Nuitrack SDK.

What Dmitry described is a mean to do preliminary feasibility study for your specific use-case (if current version of Nuitrack doesn’t cover your needs). This isn’t somehow specific to the AI version and could be proceeded for current version, so feel free to send corresponding request. Depending on specific use-case this could be already included into our roadmap or would require separate customization/fine-tuning (on a paid basis).

AI version will be backward-compatible with current version, so joint 3D positions/orientation will be accessible through the same C++/C# “modules based” API. AI version will also include additional features (faces/objects). They will be accessible through the new extendable “instance based” API (http://download.3divi.com/Nuitrack/doc/Instance_based_API.html)


#4

Dear Dmitow,

Thank you for your reply.
We will look forward to this release.
My question about data was unclear.
I want to know if we can control the learning database in order to focus is on specific case ?
For example we have golf motion data and we want that your deep-learning method learn only with these kind of data.
We can talk about that privately if you prefer.

Best regards,

PP


#5

Dear pplantard,

usually we use external customer data for validation purposes only and use our internal data sources for training. This kind of pipeline provides more guarantees on robustness of solution. At current stage of AI version development we mostly aim to cover wide range of use-cases with mainline version, rather than focus only on specific use-cases. If you have reference data for use-case of your interest, the best way to proceed is to check on our side how this input data is handled with current version. If this is the case, please feel free to contact me privately or through the feedback form at nuitrack.com


#6

Can you elaborate on the new AI faces and objects features?


#7

Nuitrack AI will get Face features - face orientation, gender age, emotions, biometric ID and several object detectors (drink, phone, cigarrete).
Btw, can you share your requirements?


#8

Thanks Dmitry. My primary concern is improved 360° skeleton tracking as well as a more persistent way to keep track of an individual skeleton. I can share a video of my application, however I would prefer not to post it on the forum. I am considering a second application which would require facial landmark tracking independent of skeleton tracking.


#9

please send video to nuitrack@3divi.com (with “Attn: Dmitry” in subject line). We use it for algorithm validation. p.s. will keep it confidential