Anyone in the Sportsbox AI Beta?

[mdl]

Sportsbox is a startup using computer vision pose estimation to do AI swing coaching. Along the lines of using modern 3D pose estimation to get something like a GEARS golf experience just from your phone and a tripod, with automated coaching built on top of the poses. Things like this:I signed up but they got enough interest they limited their beta to golf coaches. I'm interested as a player but also professionally.

[mdl]

Whoops. Messed up the first one...

Sportsbox is a startup using computer vision pose estimation to do AI swing coaching. Along the lines of using modern 3D pose estimation to get something like a GEARS golf experience just from your phone and a tripod, with automated coaching built on top of the poses. Things like this:

https://www.instagram.com/p/COyrH7UHbRQ/embed/?autoplay=1

  

I signed up but they got enough interest they limited their beta to golf coaches. I'm interested as a player but also professionally. Anyone in their beta? How's the product!?

[iacas]

I'm on the list and have talked to a few people involved with the beta. They're happy, and for what it will cost, it's pretty amazing. It's not anything close to GEARS, but it's kinda like the Mevo is to a full on FlightScope X3.

[mdl]

10 minutes ago, iacas said:

I'm on the list and have talked to a few people involved with the beta. They're happy, and for what it will cost, it's pretty amazing. It's not anything close to GEARS, but it's kinda like the Mevo is to a full on FlightScope X3.

Makes sense. Thanks Erik! Not anything close in terms of accuracy of pose estimation versus motion capture? Or other features/capabilities?

[iacas]

Just now, mdl said:

Makes sense. Thanks Erik! Not anything close in terms of accuracy of pose estimation versus motion capture? Or other features/capabilities?

Not close in anything. But it's going to be an app, so… I'm not saying something that shouldn't be a surprise. Not close in:

• the sheer number of things that can be measured
• accuracy
• price

🙂

BTW I should be in the beta in early November.

[mdl]

8 minutes ago, iacas said:

Not close in anything. But it's going to be an app, so… I'm not saying something that shouldn't be a surprise. Not close in:

• the sheer number of things that can be measured
• accuracy
• price

🙂

BTW I should be in the beta in early November.

Would love to hear your thoughts when you start using it.

[mdl]

@iacas bump. I talked with their CEO the other day. Still need deeper exploration but just from the few examples that have bubbled up to her it sounds like they've done a great job learning from cutting edge 3D pose estimation research and developing their proprietary algorithm. Most impressive number was their trail arm angle estimate at A4, where it's often totally hidden in FO. Her claim was it's right on the majority of the time and otherwise within 1-2˚ of estimates from motion capture or electromagnetic. Interested to hear how that lines up with your experience.

[iacas]

6 hours ago, mdl said:

@iacas bump. I talked with their CEO the other day. Still need deeper exploration but just from the few examples that have bubbled up to her it sounds like they've done a great job learning from cutting edge 3D pose estimation research and developing their proprietary algorithm. Most impressive number was their trail arm angle estimate at A4, where it's often totally hidden in FO. Her claim was it's right on the majority of the time and otherwise within 1-2˚ of estimates from motion capture or electromagnetic. Interested to hear how that lines up with your experience.

I don't have any experience with it. I'm not going to pay for it, and I'm not interested in starting a trial.

I'd be surprised if it was anywhere near that accurate. It's an incredibly difficult thing to do, and people have different sized bodies and parts, different sized clothing, etc. They'll be filming from some not-so-perfect angles… etc.

[Warlock]

On 12/16/2021 at 11:47 PM, iacas said:

I don't have any experience with it. I'm not going to pay for it, and I'm not interested in starting a trial.

I'd be surprised if it was anywhere near that accurate. It's an incredibly difficult thing to do, and people have different sized bodies and parts, different sized clothing, etc. They'll be filming from some not-so-perfect angles… etc.

Sorry to bump an old thread, but Dr Phil Cheetham and some other famous names seem to be actively investing in this AI technology. 

I'm still unsure how it all works using 2D video to convert into 3D.  Is it a best estimate of kinematic data measurements?

I've been viewing some of the you-tube videos and I think that Sportsbox are using Dr Phil Cheetham's 3D data that he has on tour players.  Could it be that they compare the golfer body segment measurements to a 'closest match' of a tour player and somehow plot the differences in body segment positions (over time) to determine the golfer's kinematic data?  Then create a best estimate of other kinematic data (not shown in 2D) by using the 3D information of the tour pro?

[iacas]

Those sound like the kind of questions you should ask them.

[jshots]

5 hours ago, Warlock said:

Sorry to bump an old thread, but Dr Phil Cheetham and some other famous names seem to be actively investing in this AI technology. 

I'm still unsure how it all works using 2D video to convert into 3D.  Is it a best estimate of kinematic data measurements?

I've been viewing some of the you-tube videos and I think that Sportsbox are using Dr Phil Cheetham's 3D data that he has on tour players.  Could it be that they compare the golfer body segment measurements to a 'closest match' of a tour player and somehow plot the differences in body segment positions (over time) to determine the golfer's kinematic data?  Then create a best estimate of other kinematic data (not shown in 2D) by using the 3D information of the tour pro?

I'd been wondering when we would see something like this for golf. The pace this type of machine learning is moving is incredible, and it can do things that are sometimes hard to believe.

I would bet that they film golf swings, similar to the way an end user would do it , but use a system similar to gears as the validation data to train some kind of deep learning algorithm to convert the input video into the body segments. What a time to be alive.