Publications

Flow matching models have recently emerged as an efficient alternative to diffusion, especially for text-guided image generation and editing, offering faster inference through continuous-time dynamics. However, existing flow-based editors predominantly support global or single-instruction edits and struggle with multi-instance scenarios, where multiple parts of a reference input must be edited independently without semantic interference. We identify this limitation as a consequence of globally conditioned velocity fields and joint attention mechanisms, which entangle concurrent edits. To address this issue, we introduce Instance-Disentangled Attention, a mechanism that partitions joint attention operations, enforcing binding between instance-specific textual instructions and spatial regions during velocity field estimation. We evaluate our approach on both natural image editing and a newly introduced benchmark of text-dense infographics with region-level editing instructions. Experimental results demonstrate that our approach promotes edit disentanglement and locality while preserving global output coherence, enabling single-pass, instance-level editing.

Brain tumor segmentation is a crucial task in medical imaging that involves the integrated modeling of four distinct imaging modalities to identify tumor regions accurately. Unfortunately, in real-life scenarios, the full availability of such four modalities is often violated due to scanning cost, time, and patient condition. Consequently, several deep learning models have been developed to address the challenge of brain tumor segmentation under conditions of missing imaging modalities. However, the majority of these models have been evaluated using the 2018 version of the BraTS dataset, which comprises only $285$ volumes. In this study, we reproduce and extensively analyze the most relevant models using BraTS2023, which includes 1,250 volumes, thereby providing a more comprehensive and reliable comparison of their performance. Furthermore, we propose and evaluate the adoption of Mamba as an alternative fusion mechanism for brain tumor segmentation in the presence of missing modalities. Experimental results demonstrate that transformer-based architectures achieve leading performance on BraTS2023, outperforming purely convolutional models that were instead superior in BraTS2018. Meanwhile, the proposed Mamba-based architecture exhibits promising performance in comparison to state-of-the-art models, competing and even outperforming transformers. The source code of the proposed approach is publicly released alongside the benchmark developed for the evaluation: https://github.com/AImageLab-zip/IM-Fuse.

L’Intelligenza Artificiale è già parte della nostra vita. È nei giornali che leggiamo, nelle immagini che vediamo, negli strumenti che utilizziamo ogni giorno. Eppure, per molti, continua a essere una tecnologia difficile da comprendere, sospesa tra entusiasmo, aspettative e timori. Perché oggi parliamo tanto di Intelligenza Artificiale? Perché il Deep Learning, nato decenni fa, è diventato protagonista solo negli ultimi anni? Quanto contano davvero i dati e quanti ne servono per rendere intelligente una macchina? Perché l’Europa sembra arrancare nella corsa all’IA? E possiamo davvero affidarci all’Intelligenza Artificiale per riconoscere i contenuti prodotti dall’Intelligenza Artificiale stessa? Soprattutto: dobbiamo avere paura di questa tecnologia? Attraverso dieci domande, il libro accompagna il lettore alla scoperta di una delle più profonde trasformazioni tecnologiche del nostro tempo, spiegando come funzionano queste tecnologie e perché stanno cambiando il nostro modo di vivere e lavorare. Al termine di ogni capitolo ritorna la stessa domanda: «Perché per noi è importante sapere tutto questo?». È il filo conduttore che collega la teoria alla realtà e ricorda che comprendere l’Intelligenza Artificiale non significa soltanto conoscere una nuova tecnologia, ma capire il mondo che sta contribuendo a costruire. Perché conoscere l’Intelligenza Artificiale non è più un privilegio per specialisti: è uno strumento indispensabile per chiunque voglia orientarsi nel presente e immaginare il futuro.

Generated Scalable Vector Graphics (SVG) images demand evaluation criteria tuned to their symbolic and vectorial nature – criteria that existing metrics such as FID, LPIPS, or CLIPScore fail to satisfy. In this paper, we introduce SVGauge, the first human-aligned, reference-based metric for text-to-SVG generation. SVGauge jointly measures (i) visual fidelity, obtained by extracting SigLIP image embeddings and refining them with PCA and whitening for domain alignment, and (ii) semantic consistency, captured by comparing BLIP-2-generated captions of the SVGs against the original prompts in the combined space of SBERT and TF-IDF. Evaluation on the proposed SHE benchmark shows that SVGauge attains the highest correlation with human judgments and reproduces system-level rankings of eight zero-shot LLM-based generators more faithfully than existing metrics. Our results highlight the necessity of vector-specific evaluation and provide a practical tool for benchmarking future text-to-SVG generation models.

Despite recent advances, single-image super-resolution (SR) remains challenging, especially in real-world scenarios with complex degradations. Diffusion-based SR methods, particularly those built on Stable Diffusion, leverage strong generative priors but commonly rely on text conditioning derived from semantic captioning. Such textual descriptions provide only high-level semantics and lack the spatially aligned visual information required for faithful restoration, leading to a representation gap between abstract semantics and spatially aligned visual details. To address this limitation, we propose GramSR, a one-step diffusion-based SR framework that replaces text conditioning with dense visual features extracted from the low-resolution input using a pre-trained DINOv3 encoder. GramSR adopts a three-stage LoRA architecture, where pixel-level, semantic-level, and texture-level LoRA modules are trained sequentially. The pixel-level module focuses on degradation removal using L2 loss, the semantic-level module enhances perceptual details via LPIPS and CSD losses, and the texture-level module enforces feature correlation consistency through a Gram matrix loss computed from DINOv3 features. At inference, independent guidance scales enable flexible control over degradation removal, semantic enhancement, and texture preservation. Extensive experiments on standard SR benchmarks demonstrate that GramSR consistently outperforms existing one-step diffusion-based methods, achieving superior structural fidelity and texture realism.

Model reuse offers a solution to the challenges of segmentation in biomedical imaging, where high data annotation costs remain a major bottleneck for deep learning. However, although many pre-trained models are released through challenges, model zoos, and repositories, selecting the most suitable model for a new dataset remains difficult due to the lack of reliable model ranking methods. We introduce the first black-box-compatible framework for unsupervised and source-free ranking of semantic and instance segmentation models based on the consistency of predictions under perturbations. While ranking methods have been studied for classification and a few segmentation-related approaches exist, most target-related tasks such as transferability estimation or model validation and typically rely on labelled data, feature-space access, or specific training assumptions. In contrast, our method directly addresses the repository setting and applies to both semantic and instance segmentation, for zero-shot reuse or after unsupervised domain adaptation. We evaluate the approach across a wide range of biomedical segmentation tasks in both 2D and 3D imaging, showing that our estimated rankings strongly correlate with true target-domain model performance rankings. Code is available on GitHub: https://github.com/kreshuklab/model_ranking.

Reliable estimates of cellular composition and anatomical distribution in the human brain are essential for biologically plausible circuit models. In the hippocampus, existing reconstructions rely on low-resolution (LR) data without explicit cell-type-resolved annotations, limiting quantitative maps of excitatory neurons, inhibitory interneurons, and glial cells. Using newly released 1 um/px BigBrain sections of the right hippocampus, we present CALHippo, Cellular Annotation Library for the Hippocampus, a multiscale resource for cell-type-resolved reconstruction of the human CA complex. CALHippo includes the first expert-validated, cell-level annotated dataset spanning all Cornu Ammonis (CA1-CA4) subfields with explicit three-class labels, together with a lower-resolution mesoscale cellular point-cloud map. High-resolution (HR) cell instances are obtained through a human-in-the-loop pipeline combining foundation-model-based segmentation, iterative expert correction, and model ensembling, and are classified as excitatory neurons, inhibitory interneurons, or glial cells. To extend sparse HR annotations to the full volume, we project them into the 20 um/px LR BigBrain space and use the resulting class-specific supervision maps to train a UNet-based density estimation model. The predicted density maps enable slice-by-slice inference across the full CA complex and are sampled to generate a class-resolved mesoscale cellular point cloud. Code (https://github.com/AImageLab-zip/CALHippo-Framework) and dataset (https://ditto.ing.unimore.it/calhippo) are publicly released to support reproducibility.