Alexandre Lacoste

Alors, pour l'audit de l'Allier, c'est environ 28 000 interventions sur l'année 2025. en fin de compte, ce qu'il faut comprendre, c'est que les interventions quotidiennes, les petits bobos, les malaises, les arrêts cardiaques, les noyades, tout ça, c'est ce qu'on appelle les interventions courantes. Et à cela, viennent s'ajouter les conséquences aléas climatiques. Donc, c'est le feu d'espace naturel pour la période de l'été. 'est les intempéries, quand mi-septembre va arriver, c'est la grêle, tout ça. Donc, il faut bien comprendre que les interventions, elles sont faites à flux tendu déjà toute l'année et qu'à cela s'ajoutent les événements exceptionnels 'on peut rencontrer et qu'on a notam…

completely agree with you on that. Um, and yes. The, you have to, you can create environments, a synthetic environment where, uh, you know, perhaps a fake environment where you can do this and then you can start to doing reinforcement learning to improve this. work, having worked on that for a couple of years on the UI agent side, um, it's, it works. very slow progress. And then, you know, you get this, this new agents that you got, you show that Lama, open source model can make that progress. And then you get this new closed source model that comes out and out of the box, it gets really, uh, um, uh, good performance. Um, but yeah, they, so the two world exists. So that I don't definitely d…

Yes. Um, so you, want to know why exactly, um, so, so there's, on, on one side, you have the, the SWIN and, uh, the CONVNEXT are directly with this dense tensor thing and it's able to have a more conventional approach. we see in, in, in when we measure on benchmark to VIT, the performance is a little higher. the hypothesis is that it's able to, um, uh, make better usage of this fine grain information that we see in the image. Because we're often limited by, um, the resolution. We wish we would have more pixels, but we, we don't have it. Like 0.3 meter, is the best we can get. we, we need to make the best we can out of these, um, these pixels. Um, so in CONVNEXT to be one of the, you know, i…

I think, I think it's more common in geospatial data that you're limited to your resolution, right? When you, you have common image from the web, uh, it's often high resolution and you have, like, high level semantics, uh, object that you care about. Uh, but I think if you, even from normal image, if you start caring about, many very specific details, then this patch level, uh, is going to hurt you a little bit more. And this is something we see a lot in maybe one third of geospatial application where you're, if you could have better pixels, less blurry, or high resolution, then it's going to be better. And this is where this, this patch thing is, is, it's hurting a little bit.