Search arXiv⌕ Search

arXiv subjects

Tom Kelly

Publications and source records attributed to Tom Kelly.

At least 19 recordsLinked to original sources
arXiv2026

Warping Earth Observations for better ice labeling in the Marginal Marginal Ice Zone

Multimodal satellite imagery provides complementary information for Earth Observation, but accurately combining heterogeneous sensors remains challenging in dynamic environments. Fast-changing regions, such as the Antarctic marginal ice zone, cannot fully exploit multimodal information from different satellite sensors because surface features move between image acquisitions. This spatial and temporal mismatch challenges effective perceptual grounding, violating the assumption of pixel-level correspondence that underpins most multimodal reasoning and downstream classification pipelines. Antarctic sea ice provides a challenging benchmark due to the rapid, heterogeneous drift of individual ice floes and the differing responses of sea ice to radar, visible and thermal sensing modalities. Accurate, dense supervision of sea ice remains scarce because generating pixel-wise labels requires time-consuming expert interpretation of noisy data, leading to historical reliance on coarse-resolution maritime ice charts for model training. This paper presents a novel architecture based on mutual information warping to align multi-satellite (Sentinel-1 and MODIS platforms) multimodal (visible, thermal, radar) satellite scenes. To demonstrate the approach, we introduce a sparse expert-labeled dataset of 2,088 pixel-wise annotations (7,046 expert point classifications) located at the ice-water margin interface across 43 scenes. Our results demonstrate that spatially grounding and aligning modalities prior to segmentation improves classification accuracy, and enables accurate, dense sea ice segmentation from sparse point-wise supervision.

cs.CV↗
arXiv2026

Improved bounds for the chromatic index of $k$-uniform hypergraphs

In 1997, Alon and Kim conjectured that if $H$ is a $k$-uniform $t$-simple hypergraph with maximum degree $D$ sufficiently large, then the chromatic index $\chi'(H)$ is upper bounded by $(t-1+1/t+\varepsilon)D$. Using probabilistic techniques and a nibble coloring method, we prove a general coloring theorem stating that a $k$-uniform $t$-simple hypergraph $H$ with large maximum degree $D$ satisfies $$\chi'(H) \le (b+\varepsilon)kD,$$ where $b$ is a particular parameter derived from local structural information about $H$. We use structural techniques to prove sharp upper bounds on $b$ in the 3-uniform 2-simple, and 3-uniform 3-simple cases. In particular, we deduce as a corollary that for sufficiently large $D$, every 3-uniform 2-simple and 3-simple hypergraph of maximum degree at most $D$ has chromatic index at most $2.3581D$ and $2.6791D$, respectively.

math.CO↗
arXiv2025

Advancing the R\"{o}dl Nibble: New bounds on matchings and the list chromatic index of hypergraphs

Let $H$ be a $(k+1)$-uniform hypergraph which is nearly $D$-regular, such that any set of $i$ vertices is contained in at most $D_i$ edges of $H$ for each $i = 2, 3, \dots, k+1$. Influential results of Pippenger and of Frankl and R\"odl show that the \textit{R\"odl Nibble} -- a probabilistic procedure which iteratively constructs a matching in small bits -- can produce an almost-perfect matching in $H$, provided $D_2$ is much smaller than $D$. The quantitative aspects of this result were sharpened by several authors, with the previously best-known result due to Vu, whose result takes more of the codegree sequence $D_2, \dots, D_{k+1}$ into account. We improve Vu's result, by showing the R\"odl Nibble can ``exhaust'' the full codegree sequence up to one of several natural bottlenecks, even tolerating extensive ``clustering'' of codegree values. Up to a subpolynomial error term, we believe our result to be the optimal usage of pure nibble methodology. We also show that our matching can be taken to be ``pseudorandom'' with respect to a set of weight functions on $V(H)$, and we use this result to derive other hypergraph matching results in partite settings, including a new bound on the list chromatic index which implies the best-known result of Molloy and Reed up to the error term, and is stronger when the hypergraph is not close to linear, i.e.\ $D_2=\omega(1)$. We also apply our results to obtain improved bounds on almost-spanning structures in Latin squares and designs, and the maximum diameter of a simplicial complex.

math.CO↗
arXiv2025

Entropy Bounds for Perfect Matchings in Bipartite Hypergraphs

A hypergraph is \textit{bipartite with bipartition $(A, B)$} if every edge has exactly one vertex in $A$, and a matching in such a hypergraph is \textit{$A$-perfect} if it saturates every vertex in $A$. We prove an upper bound on the number of $A$-perfect matchings in uniform hypergraphs with small maximum codegree. Using this result, we prove that there exist order-$n$ Latin squares with at most $(n/e^{2.117})^n$ transversals when $n$ is odd and $n \equiv 0\pmod 3$. We also show that $k$-uniform $D$-regular hypergraphs on $n$ vertices have at most $((1+o(1))q/e^k)^{Dn/k}$ proper $q$-edge-colorings when $q = (1+o(1))D$ and the maximum codegree is $o(q)$.

math.CO↗
arXiv2024

A Short Proof of the Existence of $K_q^r$-absorbers

We codify a short self-contained proof of the existence of $K_q^r$-absorbers implicit in Keevash's original proof of the Existence Conjecture. Combining this with the work of the first and third authors in yields a proof of the Existence Conjecture for Combinatorial Designs that is not reliant on the construction of $K_q^r$-absorbers by Glock, K\"uhn, Lo, and Osthus.

math.CO↗
arXiv2024

Improving the Caro-Wei bound and applications to Tur\'{a}n stability

We prove that if $G$ is a graph and $f(v) \leq 1/(d(v) + 1/2)$ for each $v\in V(G)$, then either $G$ has an independent set of size at least $\sum_{v\in V(G)}f(v)$ or $G$ contains a clique $K$ such that $\sum_{v\in K}f(v) > 1$. This result implies that for any $\sigma \leq 1/2$, if $G$ is a graph and every clique $K\subseteq V(G)$ has at most $(1 - \sigma)(|K| - \sigma)$ simplicial vertices, then $\alpha(G) \geq \sum_{v\in V(G)} 1 / (d(v) + 1 - \sigma)$. Letting $\sigma = 0$ implies the famous Caro-Wei Theorem, and letting $\sigma = 1/2$ implies that if fewer than half of the vertices in each clique of $G$ are simplicial, then $\alpha(G) \geq \sum_{v\in V(G)}1/(d(v) + 1/2)$, which is tight for the 5-cycle. When applied to the complement of a graph, this result implies the following new Tur\' an stability result. If $G$ is a $K_{r + 1}$-free graph with more than $(1 - 1/r)n^2/2 - n/4$ edges, then $G$ contains an independent set $I$ such that at least half of the vertices in $I$ are complete to $G - I$. Applying this stability result iteratively provides a new proof of the stability version of Tur\' an's Theorem in which $K_{r + 1}$-free graphs with close to the extremal number of edges are $r$-partite.

math.CO↗
arXiv2024

On the existence of reflecting $n$-queens configurations

In 1967, Klarner proposed a problem concerning the existence of reflecting $n$-queens configurations. The problem considers the feasibility of placing $n$ mutually non-attacking queens on the reflecting chessboard, an $n\times n$ chessboard with a $1\times n$ "reflecting strip" of squares added along one side of the board. A queen placed on the reflecting chessboard can attack the squares in the same row, column, and diagonal, with the additional feature that its diagonal path can be reflected via the reflecting strip. Klarner noted the equivalence of this problem to a number theory problem proposed by Slater, which asks: for which $n$ is it possible to pair up the integers 1 through $n$ with the integers $n+1$ through $2n$ such that no two of the sums or differences of the $n$ pairs of integers are the same. We prove the existence of reflecting $n$-queens configurations for all sufficiently large $n$, thereby resolving both Slater's and Klarner's questions for all but a finite number of integers.

math.CO↗
arXiv2024

Clique Decompositions in Random Graphs via Refined Absorption

We prove that if $p\ge n^{-\frac{1}{3}+\beta}$ for some $\beta > 0$, then asymptotically almost surely the binomial random graph $G(n,p)$ has a $K_3$-packing containing all but at most $n + O(1)$ edges. Similarly, we prove that if $d \ge n^{\frac{2}{3}+\beta}$ for some $\beta > 0$ and $d$ is even, then asymptotically almost surely the random $d$-regular graph $G_{n,d}$ has a triangle decomposition provided $3 \mid d \cdot n$. We also show that $G(n,p)$ admits a fractional $K_3$-decomposition for such a value of $p$. We prove analogous versions for a $K_q$-packing of $G(n,p)$ with $p\ge n^{-\frac{1}{q+0.5}+\beta}$ and leave of $(q-2)n+O(1)$ edges, for $K_q$-decompositions of $G_{n,d}$ with $(q-1)~|~d$ and $d\ge n^{1-\frac{1}{q+0.5}+\beta}$ provided $q\mid d\cdot n$, and for fractional $K_q$-decompositions.

math.CO↗
arXiv2023

Subsquares in random Latin squares and rectangles

A $k \times n$ partial Latin rectangle is \textit{$C$-sparse} if the number of nonempty entries in each row and column is at most $C$ and each symbol is used at most $C$ times. We prove that the probability a uniformly random $k \times n$ Latin rectangle, where $k < (1/2 - \alpha)n$, contains a $\beta n$-sparse partial Latin rectangle with $\ell$ nonempty entries is $(\frac{1 \pm \varepsilon}{n})^\ell$ for sufficiently large $n$ and sufficiently small $\beta$. Using this result, we prove that a uniformly random order-$n$ Latin square asymptotically almost surely has no Latin subsquare of order greater than $c\sqrt{n\log n}$ for an absolute constant $c$.

math.CO↗
arXiv2023

FacadeNet: Conditional Facade Synthesis via Selective Editing

We introduce FacadeNet, a deep learning approach for synthesizing building facade images from diverse viewpoints. Our method employs a conditional GAN, taking a single view of a facade along with the desired viewpoint information and generates an image of the facade from the distinct viewpoint. To precisely modify view-dependent elements like windows and doors while preserving the structure of view-independent components such as walls, we introduce a selective editing module. This module leverages image embeddings extracted from a pre-trained vision transformer. Our experiments demonstrated state-of-the-art performance on building facade generation, surpassing alternative methods.

cs.CV↗
arXiv2023

WinSyn: A High Resolution Testbed for Synthetic Data

We present WinSyn, a unique dataset and testbed for creating high-quality synthetic data with procedural modeling techniques. The dataset contains high-resolution photographs of windows, selected from locations around the world, with 89,318 individual window crops showcasing diverse geometric and material characteristics. We evaluate a procedural model by training semantic segmentation networks on both synthetic and real images and then comparing their performances on a shared test set of real images. Specifically, we measure the difference in mean Intersection over Union (mIoU) and determine the effective number of real images to match synthetic data's training performance. We design a baseline procedural model as a benchmark and provide 21,290 synthetically generated images. By tuning the procedural model, key factors are identified which significantly influence the model's fidelity in replicating real-world scenarios. Importantly, we highlight the challenge of procedural modeling using current techniques, especially in their ability to replicate the spatial semantics of real-world scenarios. This insight is critical because of the potential of procedural models to bridge to hidden scene aspects such as depth, reflectivity, material properties, and lighting conditions.

cs.CV↗
arXiv2023

Optimal spread for spanning subgraphs of Dirac hypergraphs

Let $G$ and $H$ be hypergraphs on $n$ vertices, and suppose $H$ has large enough minimum degree to necessarily contain a copy of $G$ as a subgraph. We give a general method to randomly embed $G$ into $H$ with good "spread". More precisely, for a wide class of $G$, we find a randomised embedding $f\colon G\hookrightarrow H$ with the following property: for every $s$, for any partial embedding $f'$ of $s$ vertices of $G$ into $H$, the probability that $f$ extends $f'$ is at most $O(1/n)^s$. This is a common generalisation of several streams of research surrounding the classical Dirac-type problem. For example, setting $s=n$, we obtain an asymptotically tight lower bound on the number of embeddings of $G$ into $H$. This recovers and extends recent results of Glock, Gould, Joos, K\"uhn, and Osthus and of Montgomery and Pavez-Sign\'e regarding enumerating Hamilton cycles in Dirac hypergraphs. Moreover, using the recent developments surrounding the Kahn--Kalai conjecture, this result implies that many Dirac-type results hold robustly, meaning $G$ still embeds into $H$ after a random sparsification of its edge set. This allows us to recover a recent result of Kang, Kelly, K\"uhn, Osthus, and Pfenninger and of Pham, Sah, Sawhney, and Simkin for perfect matchings, and obtain novel results for Hamilton cycles and factors in Dirac hypergraphs. Notably, our randomised embedding algorithm is self-contained and does not require Szemer\'edi's regularity lemma or iterative absorption.

math.CO↗
arXiv2023

The independence ratio of 4-cycle-free planar graphs

We prove that every $n$-vertex planar graph $G$ with no triangle sharing an edge with a 4-cycle has independence ratio $n/\alpha(G) \leq 4 - \varepsilon$ for $\varepsilon = 1/30$. This result implies that the same bound holds for 4-cycle-free planar graphs and planar graphs with no adjacent triangles and no triangle sharing an edge with a 5-cycle. For the latter case we strengthen the bound to $\varepsilon = 2/9$.

math.CO↗
arXiv2022

Perfect matchings in random sparsifications of Dirac hypergraphs

For all integers $n \geq k > d \geq 1$, let $m_{d}(k,n)$ be the minimum integer $D \geq 0$ such that every $k$-uniform $n$-vertex hypergraph $\mathcal H$ with minimum $d$-degree $\delta_{d}(\mathcal H)$ at least $D$ has an optimal matching. For every fixed integer $k \geq 3$, we show that for $n \in k \mathbb{N}$ and $p = \Omega(n^{-k+1} \log n)$, if $\mathcal H$ is an $n$-vertex $k$-uniform hypergraph with $\delta_{k-1}(\mathcal H) \geq m_{k-1}(k,n)$, then a.a.s.\ its $p$-random subhypergraph $\mathcal H_p$ contains a perfect matching. Moreover, for every fixed integer $d < k$ and $\gamma > 0$, we show that the same conclusion holds if $\mathcal H$ is an $n$-vertex $k$-uniform hypergraph with $\delta_d(\mathcal H) \geq m_{d}(k,n) + \gamma\binom{n - d}{k - d}$. Both of these results strengthen Johansson, Kahn, and Vu's seminal solution to Shamir's problem and can be viewed as ``robust'' versions of hypergraph Dirac-type results. In addition, we also show that in both cases above, $\mathcal H$ has at least $\exp((1-1/k)n \log n - \Theta (n))$ many perfect matchings, which is best possible up to an $\exp(\Theta(n))$ factor.

math.CO↗
arXiv2022

Large-Scale Auto-Regressive Modeling Of Street Networks

We present a novel generative method for the creation of city-scale road layouts. While the output of recent methods is limited in both size of the covered area and diversity, our framework produces large traversable graphs of high quality consisting of vertices and edges representing complete street networks covering 400 square kilometers or more. While our framework can process general 2D embedded graphs, we focus on street networks due to the wide availability of training data. Our generative framework consists of a transformer decoder that is used in a sliding window manner to predict a field of indices, with each index encoding a representation of the local neighborhood. The semantics of each index is determined by a dictionary of context vectors. The index field is then input to a decoder to compute the street graph. Using data from OpenStreetMap, we train our system on whole cities and even across large countries such as the US, and finally compare it to the state of the art.

cs.LG↗
arXiv2022

Thresholds for Latin squares and Steiner triple systems: Bounds within a logarithmic factor

We prove that for $n \in \mathbb N$ and an absolute constant $C$, if $p \geq C\log^2 n / n$ and $L_{i,j} \subseteq [n]$ is a random subset of $[n]$ where each $k\in [n]$ is included in $L_{i,j}$ independently with probability $p$ for each $i, j\in [n]$, then asymptotically almost surely there is an order-$n$ Latin square in which the entry in the $i$th row and $j$th column lies in $L_{i,j}$. The problem of determining the threshold probability for the existence of an order-$n$ Latin square was raised independently by Johansson, by Luria and Simkin, and by Casselgren and H{\"a}ggkvist; our result provides an upper bound which is tight up to a factor of $\log n$ and strengthens the bound recently obtained by Sah, Sawhney, and Simkin. We also prove analogous results for Steiner triple systems and $1$-factorizations of complete graphs, and moreover, we show that each of these thresholds is at most the threshold for the existence of a $1$-factorization of a nearly complete regular bipartite graph.

math.CO↗
arXiv2022

Projective Urban Texturing

This paper proposes a method for automatic generation of textures for 3D city meshes in immersive urban environments. Many recent pipelines capture or synthesize large quantities of city geometry using scanners or procedural modeling pipelines. Such geometry is intricate and realistic, however the generation of photo-realistic textures for such large scenes remains a problem. We propose to generate textures for input target 3D meshes driven by the textural style present in readily available datasets of panoramic photos capturing urban environments. Re-targeting such 2D datasets to 3D geometry is challenging because the underlying shape, size, and layout of the urban structures in the photos do not correspond to the ones in the target meshes. Photos also often have objects (e.g., trees, vehicles) that may not even be present in the target geometry. To address these issues we present a method, called Projective Urban Texturing (PUT), which re-targets textural style from real-world panoramic images to unseen urban meshes. PUT relies on contrastive and adversarial training of a neural architecture designed for unpaired image-to-texture translation. The generated textures are stored in a texture atlas applied to the target 3D mesh geometry. To promote texture consistency, PUT employs an iterative procedure in which texture synthesis is conditioned on previously generated, adjacent textures. We demonstrate both quantitative and qualitative evaluation of the generated textures.

cs.CV↗