Go to Villagebet.com.au for free horse racing tips - Click here now
Forum Home Forum Home > Horse Breeding - Public Forums > Conformation
  New Posts New Posts RSS Feed - Markerless Biomechanics
  FAQ FAQ  Forum Search   Events   Register Register  Login Login


Thoroughbred Village Home Page. For village news, follow @TBVillage on Twitter. For horseracing tips, follow @Villagebet on Twitter. To contact the Mayor by email: Click Here.

Markerless Biomechanics
 Post Reply Post Reply
Author
Message Reverse Sort Order
  Topic Search Topic Search  Topic Options Topic Options
GAJ View Drop Down
Champion
Champion
Avatar

Joined: 13 Apr 2011
Location: Alstonville,NSW
Status: Offline
Points: 4410 		Post Options Post Options   Thanks (0) Thanks(0)   Quote GAJ Quote  Post ReplyReply Direct Link To This Post Topic: Markerless Biomechanics
    Posted: 19 Apr 2020 at 5:08pm
Amazing, Thank you, Then would come the gene test I'm guessing.
Back to Top
Red Hare View Drop Down
Champion
Champion


Joined: 12 Jan 2015
Location: Victoria
Status: Offline
Points: 4233 		Post Options Post Options   Thanks (0) Thanks(0)   Quote Red Hare Quote  Post ReplyReply Direct Link To This Post Posted: 19 Apr 2020 at 2:01pm
GAJ, just in case old mate doesn’t come back, let me help.

The aim of the game here is to create a program that can identify good racehorses using digital inputs - i.e. what’s available via online catalogues. In this case, the video that shows the horse at a walk.

The study is suggesting that pre-training the artificial intelligence will yield better results than simply turning it on & allowing it to train itself from scratch.
Back to Top
GAJ View Drop Down
Champion
Champion
Avatar

Joined: 13 Apr 2011
Location: Alstonville,NSW
Status: Offline
Points: 4410 		Post Options Post Options   Thanks (0) Thanks(0)   Quote GAJ Quote  Post ReplyReply Direct Link To This Post Posted: 13 Apr 2020 at 7:08pm
Originally posted by brogers brogers wrote:

Here is the paper that has come out of the work....

Pretraining boosts out-of-domain robustness for pose estimation - https://arxiv.org/abs/1909.11229

 
Hi Brogers, can you explain this in simple terms please.
Back to Top
Red Hare View Drop Down
Champion
Champion


Joined: 12 Jan 2015
Location: Victoria
Status: Offline
Points: 4233 		Post Options Post Options   Thanks (1) Thanks(1)   Quote Red Hare Quote  Post ReplyReply Direct Link To This Post Posted: 29 Mar 2020 at 2:00pm
Challenging times for the industry, and yet a great opportunity for the practical applications of this research.
Back to Top
brogers View Drop Down
Champion
Champion
Avatar

Joined: 20 Feb 2007
Location: Lexington, KY
Status: Offline
Points: 1106 		Post Options Post Options   Thanks (1) Thanks(1)   Quote brogers Quote  Post ReplyReply Direct Link To This Post Posted: 03 Oct 2019 at 9:58pm
Here is the paper that has come out of the work....

Pretraining boosts out-of-domain robustness for pose estimation - https://arxiv.org/abs/1909.11229
Abel Tasman...Keen Ice...Divisidero....Verrazano...Catchy...Carrick...Ivictory...
Back to Top
brogers View Drop Down
Champion
Champion
Avatar

Joined: 20 Feb 2007
Location: Lexington, KY
Status: Offline
Points: 1106 		Post Options Post Options   Thanks (1) Thanks(1)   Quote brogers Quote  Post ReplyReply Direct Link To This Post Posted: 23 Jan 2019 at 7:04am
I managed to traina neural network to properly place markers on a horse at the walk. I trained the network on 50 different horses using different light, horse color, ages, etc and the average error rate was just 9 pixels so it was very accurate in learning where to place the markers.

Here is a video I took of the Breeders Cup winner Accelerate on the left with the markers placed by the algorithm on the right.


And here is one that I downloaded from Twitter of Zoustar and had the algorithm mark up.
Abel Tasman...Keen Ice...Divisidero....Verrazano...Catchy...Carrick...Ivictory...
Back to Top
 Post Reply Post Reply

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.05
Copyright ©2001-2022 Web Wiz Ltd.

This page was generated in 0.125 seconds.