Our first efforts are paying off!

We are nearly half of the project, with more than 22000 annotated images. By curiosity, we tried to mix the images from all species in one dataset, with Yolo making no distinction between pigs, goats or cattle.  It might seem like a crazy idea, but the background for all these species is not so different, so it might be seen as detecting all the elements that are not part of this background.

It actually feels like a great idea when we see the results, here is a video, with examples from several species:

Thank you all for the sharing of data, and the diversity, multiple species and breeds, angle of view, IR and RGB, drones etc … everything is mixed into one single dataset, to increase as much as possible the capacity of Yolo to generalize: be efficient on images that haven’t been seem so far.

And this is our next challenge! Although the previous video is great, it is run on setup that have been included into our training data. Now, we want the same precision, and even more, on new images!

Don’t hesitate to spread the word to folks around you, let’s connect mathieu.bonneau@inrae.fr!

For those who want to make first tests with the latest Yolo, you can download one at the bottom of this page. It’s the Yolov11 nano version, while the video is with the large … Just to make you a little impatient! The full dataset and the best YOLO models will be shared at the end of the project!