arXiv subjects

Rogier Brussee

Publications and source records attributed to Rogier Brussee.

7 recordsLinked to original sources

arXiv2025

DistillFSS: Synthesizing Few-Shot Knowledge into a Lightweight Segmentation Model

Cross-Domain Few-Shot Semantic Segmentation (CD-FSS) seeks to segment unknown classes in unseen domains using only a few annotated examples. This setting is inherently challenging: source and target domains exhibit substantial distribution shifts, label spaces are disjoint, and support images are scarce--making standard episodic methods unreliable and computationally demanding at test time. To address these constraints, we propose DistillFSS, a framework that embeds support-set knowledge directly into a model's parameters through a teacher--student distillation process. By internalizing few-shot reasoning into a dedicated layer within the student network, DistillFSS eliminates the need for support images at test time, enabling fast, lightweight inference, while allowing efficient extension to novel classes in unseen domains through rapid teacher-driven specialization. Combined with fine-tuning, the approach scales efficiently to large support sets and significantly reduces computational overhead. To evaluate the framework under realistic conditions, we introduce a new CD-FSS benchmark spanning medical imaging, industrial inspection, and remote sensing, with disjoint label spaces and variable support sizes. Experiments show that DistillFSS matches or surpasses state-of-the-art baselines, particularly in multi-class and multi-shot scenarios, while offering substantial efficiency gains. The code is available at https://github.com/pasqualedem/DistillFSS.

cs.CV↗

arXiv2025

Matching-Based Few-Shot Semantic Segmentation Models Are Interpretable by Design

Few-Shot Semantic Segmentation (FSS) models achieve strong performance in segmenting novel classes with minimal labeled examples, yet their decision-making processes remain largely opaque. While explainable AI has advanced significantly in standard computer vision tasks, interpretability in FSS remains virtually unexplored despite its critical importance for understanding model behavior and guiding support set selection in data-scarce scenarios. This paper introduces the first dedicated method for interpreting matching-based FSS models by leveraging their inherent structural properties. Our Affinity Explainer approach extracts attribution maps that highlight which pixels in support images contribute most to query segmentation predictions, using matching scores computed between support and query features at multiple feature levels. We extend standard interpretability evaluation metrics to the FSS domain and propose additional metrics to better capture the practical utility of explanations in few-shot scenarios. Comprehensive experiments on FSS benchmark datasets, using different models, demonstrate that our Affinity Explainer significantly outperforms adapted standard attribution methods. Qualitative analysis reveals that our explanations provide structured, coherent attention patterns that align with model architectures and and enable effective model diagnosis. This work establishes the foundation for interpretable FSS research, enabling better model understanding and diagnostic for more reliable few-shot segmentation systems. The source code is publicly available at https://github.com/pasqualedem/AffinityExplainer.

cs.CV↗

arXiv2023

Fuzzy Logic and Markov Kernels

Fuzzy logic is a way to argue with boolean predicates for which we only have a confidence value between 0 and 1 rather than a well defined truth value. It is tempting to interpret such a confidence as a probability. We use Markov kernels, parametrised probability distributions, to do just that. As a consequence we get general fuzzy logic connectives from probabilistic computations on products of the booleans, stressing the importance of joint confidence functions. We discuss binary logic connectives in detail and recover the "classic" fuzzy connectives as bounds for the confidence for general connectives. We push multivariable logic formulas as far as being able to define fuzzy quantifiers and estimate the confidence.

cs.LO↗

arXiv2006

Probing Moduli Spaces of Sheaves with Donaldson and Seiberg Witten Invariants

We use Donaldson invariants of regular surfaces with p_g >0 to make quantitative statements about modulispaces of stable rank 2 sheaves. We give two examples: a quantitative existence theorem for stable bundles, and a computation of the rank of the canonical holomorphic two forms on the moduli space. The results are in some sense dual to the Donaldson and O'Grady non vanishing theorems because they use the Donaldson series of the surface as _input_. Results in purely algebraic geometric terms can be obtained by using the explicit form of the Donaldson series of the surface. The Donaldson series are easy to compute using the Seiberg Witten invariants and the Witten conjecture which has recently been rigorously proved by Feehan and Leness.

math.AG↗

arXiv1995

The canonical class and the $C^\infty$ properties of Kähler surfaces

We give a self contained proof using Seiberg Witten invariants that for Kähler surfaces with non negative Kodaira dimension (including those with $p_g = 0$) the canonical class of the minimal model and the $(-1)$-curves, are oriented diffeomorphism invariants up to sign. This implies that the Kodaira dimension is determined by the underlying differentiable manifold (Van de Ven Conjecture). We use a set up that replaces generic metrics by the construction of a localised Euler class of an infinite dimensional bundle with a Fredholm section. This allows us to compute the Seiberg Witten invariants of all elliptic surfaces with excess intersection theory. We then reprove that the multiplicities of the elliptic fibration are determined by the underlying oriented manifold, and that the plurigenera of a surface are oriented diffeomorphism invariants.

alg-geom↗

arXiv1994

Blow-up formulas for (-2)-spheres

Let $X$ be a simply connected 4-manifold containing a $(-1)$-sphere $e$. Fintushel and Stern prove that $$ D_c(\exp(te)) = D_c(B(t)) on e^\perp if c\cdot e is even, $$ $$ D_c(\exp(te)) = D_{c-e}(S(t)) on e^\perp if c \cdot e is odd, $$ for some universal series $B(t),S(t) \in \Q[x][[t]]$ with $x$ the class of a point. We show that their method can easily be extended to $(-2)$-spheres $τ$ to give blow up formulas like $$ D_c(\exp(tτ)) = D_c(B^2(t) + S^2(t)/2 τ^2) on τ^perp if c\cdot τis even. $$

dg-ga↗

arXiv1992

On the $(-1)$-curve conjecture of Friedman and Morgan

Main difference with previous version: we prove that every differentiably embedded sphere with self intersection $-1$ in a simply connected algebraic surface with $p_g >0$ is homologous to a $(-1)$-curve if $|K_{\min}|$ contains a smooth irreducible curve of genus at least 2 and $p_g$ is even or $K_{\min}^2 \not\equiv 7 \pmod8$ (here $K_{\min}$ is the canonical class of the minimal model).

alg-geom↗