2D to 3D AI-driven real-time form biomechanics app

[dekuscrub]

Disc speed/wrist speed might be good. I think the tip of the index finger/last point of contact would be ideal but I suspect that's going to be a harder problem.

Yeah, that would probably be difficult. The model does do hand tracking, but, given the way it sometimes loses track of arms obscured by the body, I imagine that trying to track an index finger under a disc would really put the model on the struggle bus.

Brilliant solution to speed inference. And in fact, you could use some of the Gatekeeper Media coverage if there are any at 90 degrees to the tee to check the estimate against what their radar measurement says

This is such a great idea! I'll have to check it out. Right now, I have no idea how accurate these estimates are. My current validation method is "Eh, I guess that seems reasonable."

 

[Sheep]

Disc speed/wrist speed might be good. I think the tip of the index finger/last point of contact would be ideal but I suspect that's going to be a harder problem.

Brilliant solution to speed inference. And in fact, you could use some of the Gatekeeper Media coverage if there are any at 90 degrees to the tee to check the estimate against what their radar measurement says

I am buddies with Derrick if we need to try and get some other footage from him.
He's really hip on stuff like this as well, so he will be highly likely to help, especially if its for academic purposes. Just makes sure whoever uses the video watermarks gatekeeper in it to give them credit, that will make him happy.

So basically I'm saying I can probably get the base files, not the youtube messups.
Cause youtube processing does mess with things.

They going into off season also, so it will be far far easier to get footage from him also.

 

[dekuscrub]

I am buddies with Derrick if we need to try and get some other footage from him.
He's really hip on stuff like this as well, so he will be highly likely to help, especially if its for academic purposes. Just makes sure whoever uses the video watermarks gatekeeper in it to give them credit, that will make him happy.

So basically I'm saying I can probably get the base files, not the youtube messups.
Cause youtube processing does mess with things.

They going into off season also, so it will be far far easier to get footage from him also.

This would be awesome! Having high quality footage parallel to the teepad + disc speeds would be huge for testing and validating these models. (Even better would be having drive distances. I know Central Coast tracks that. But not sure about Gatekeeper.)

But, yeah, if you are somehow able to get the raw footage, I'd be eternally grateful!

 

[Sheep]

This would be awesome! Having high quality footage parallel to the teepad + disc speeds would be huge for testing and validating these models. (Even better would be having drive distances. I know Central Coast tracks that. But not sure about Gatekeeper.)

But, yeah, if you are somehow able to get the raw footage, I'd be eternally grateful!

Gimmie some time.
The way I have to get ahold of him is through someone else. Cause I don't have his personal number.

 

[dekuscrub]

I haven't been able to run the model on new videos recently. So, I've been working on simplifying the code, implementing the new features, and improving the visuals.

Here's our boy GG again:

 

[Brychanus]

I haven't been able to run the model on new videos recently. So, I've been working on simplifying the code, implementing the new features, and improving the visuals.

Here's our boy GG again:

Starting to look GREAT!

 

[dekuscrub]

Haven't had been able to put a substantial amount of work into this for awhile. But I did recently make a video for Albert Tamm's 710 feet throw at the USDGC distance competition.

The code is still a convoluted mess, so my current goal is to get it into a more usable state that can turn videos quicker. Ideally, I'd like to get this running on a phone or something. But that's still a long ways off.