HorseTechie are working with a team at the Nimbus centre at CIT to develop a mobile app that will help take good sideways on photographs of horses. This will make it easier to take pictures that show off your horse well by directing you to position the horse and hold the camera at the best angle. Pictures taken over time with this app will show how your horse has changed condition and shape.
But in order to do this, we have to train the computer to recognise the sideways on profile of a horse, and for that we need LOTS of pictures. If you can help, please send images by email to firstname.lastname@example.org, one image per email and put the horses name in the subject.
At the moment we just need side on images of a horse standing, with or without rider, with no other horse in the picture. We need both pictures with the horse standing in a good posture and less good. Pictures like these:
The images you supply will be used to “train” a computer to recognise horse images. We may display your images on our website but will not link them to you without your permission. We will save your email address for occassional updates but will not share it with other organisations.
We use a method called Supervised Learning, which takes place in three steps.
First, we need to collect a lot of images of horses and a lot of images that don’t have a horse in them. These images are tagged Horse or Not Horse.
Then we take two thirds of these images or Horses and Not Horses and let the computer develop it’s own model.
Finally, we improve the model by getting the computer to test itself with the remaining third of the data and and it will change its model when it get’s the wrong answer.
If the model is accurate enough for our purposes we are done, but if it’s not yet accurate enough, we need to give it more examples of Horses and Not Horses.
The next step is to recognise when the horse is standing in a good posture. The process is the same but we need many samples that have been judged to be good and here we will need human help to tag the images as Good Stance and Not Good Stance.
Alex Vakaloudis - Project Manager
Guillame Le Bras - Developer
Elizabeth Levert - Developer