Early 2025 update

LIB has been running for 4 months now, Céline and Erwan have already did a great job, let's have a look!

The dataset is currently composed of 10 428 images for small ruminants (sheep and goats), 3 823 for pigs, and 2 173 for cows.

Thank you to all the contributors, but I hope there is still many to come! We tried to maximize the diversity of pictures, different breeds, environments, farming systems, camera type, and angle of view !
Here are some examples of the different set-up we have so far.

For Pigs:

For Small-Ruminants:

And for Cows:

The dataset is used to train Yolo type neural network in order to automatically detect animals and their head. Our Yolo is also able to estimate posture, between Standing, Lying and Transitioning.
Detection of animals and animals head is then used to have the location of the animal inside the pen. Then, you can derive several statistics, such as characterize spatial use, access to particular area, postural/activity time-budget and possibly finer behaviors. But this will be the topic a a next post !

For the most impatient, you can download a Yolov11 detector for small ruminants at the end of this page. It is the nano version of Yolo, trained for 80 epochs. Obviously, it’s still not optimal and can be improved using a larger version of Yolo, more epochs … and by collecting more pictures and that’s up to you!
Let’s collaborate together on this!