Matteo Muratori

Matteo Muratori, Joël Seytre

While state-of-the-art background removal models excel at realistic imagery, they frequently underperform in specialized domains such as anime-style content, where complex features like hair and transparency present unique challenges. To address this limitation, we collected and annotated a custom dataset of 1,228 high-quality anime images of characters and objects, and fine-tuned the open-sourced BiRefNet model on this dataset. This resulted in marked improvements in background removal accuracy for anime-style images, increasing from 95.3% to 99.5% for our newly introduced Pixel Accuracy metric. We are open-sourcing the code, the fine-tuned model weights, as well as the dataset at: https://github.com/MatteoKartoon/BiRefNet.

Matteo Muratori, Giuseppe Savaré

This is the first of a series of papers devoted to a thorough analysis of the class of gradient flows in a metric space $(X,\mathsf{d})$ that can be characterized by Evolution Variational Inequalities. We present new results concerning the structural properties of solutions to the $\mathrm{EVI}$ formulation, such as contraction, regularity, asymptotic expansion, precise energy identity, stability, asymptotic behaviour and their link with the geodesic convexity of the driving functional. Under the crucial assumption of the existence of an $\mathrm{EVI}$ gradient flow, we will also prove two main results: the equivalence with the De Giorgi variational characterization of curves of maximal slope and the convergence of the Minimizing Movement-JKO scheme to the $\mathrm{EVI}$ gradient flow, with an explicit and uniform error estimate of order $1/2$ with respect to the step size, independent of any geometric hypothesis (such as upper or lower curvature bounds) on $\mathsf{d}$. In order to avoid any compactness assumption, we will also introduce a suitable relaxation of the Minimizing Movement algorithm obtained by the Ekeland variational principle, and we will prove its uniform convergence as well.