Publications

Federated continual learning (FCL) enables models to learn new tasks across multiple distributed clients, protecting privacy and without forgetting previously acquired knowledge. However, current methods face challenges balancing performance, privacy preservation, and communication efficiency. We introduce a Distributed Online LoRA for Federated INcremental learning methodDOLFIN, a novel approach combining Vision Transformers with low-rank adapters designed to efficiently and stably learn new tasks in federated environments. Our method leverages LoRA for minimal communication overhead and incorporates Dual Gradient Projection Memory (DualGPM) to prevent forgetting. Evaluated on CIFAR-100, ImageNet-R, ImageNet-A, and CUB-200 under two Dirichlet heterogeneity settings,DOLFINconsistently surpasses six strong baselines in final average accuracy while matching their memory footprint. Orthogonal low-rank adapters offer an effective and scalable solution for privacy-preserving continual learning in federated settings.

Tool learning with foundation models aims to endow AI systems with the ability to invoke external resources — such as APIs, computational utilities, and specialized models — to solve complex tasks beyond the reach of standalone language generation. While recent advances in Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) have expanded their reasoning and perception capabilities, existing tool-use methods are predominantly limited to text-only inputs and closed-world settings. Consequently, they struggle to interpret multimodal user instructions and cannot generalize to tools unseen during training. In this work, we introduce RaTA-Tool, a novel framework for open-world multimodal tool selection. Rather than learning direct mappings from user queries to fixed tool identifiers, our approach enables an MLLM to convert a multimodal query into a structured task description and subsequently retrieve the most appropriate tool by matching this representation against semantically rich, machine-readable tool descriptions. This retrieval-based formulation naturally supports extensibility to new tools without retraining. To further improve alignment between task descriptions and tool selection, we incorporate a preference-based optimization stage using Direct Preference Optimization (DPO). To support research in this setting, we also introduce the first dataset for open-world multimodal tool use, featuring standardized tool descriptions derived from Hugging Face model cards. Extensive experiments demonstrate that our approach significantly improves tool-selection performance, particularly in open-world, multimodal scenarios.

Trajectory forecasting is crucial for video surveillance analytics, as it enables the anticipation of future movements for a set of agents, e.g., basketball players engaged in intricate interactions with long-term intentions. Deep generative models offer a natural learning approach for trajectory forecasting, yet they encounter difficulties in achieving an optimal balance between sampling fidelity and diversity. We address this challenge by leveraging Vector Quantized Variational Autoencoders (VQ-VAEs), which utilize a discrete latent space to tackle the issue of posterior collapse. Specifically, we introduce an instance-based codebook that allows tailored latent representations for each example. In a nutshell, the rows of the codebook are dynamically adjusted to reflect contextual information (i.e., past motion patterns extracted from the observed trajectories). In this way, the discretization process gains flexibility, leading to improved reconstructions. Notably, instance-level dynamics are injected into the codebook through low-rank updates, which restrict the customization of the codebook to a lower dimension space. The resulting discrete space serves as the basis of the subsequent step, which regards the training of a diffusion-based predictive model. We show that such a two-fold framework, augmented with instance-level discretization, leads to accurate and diverse forecasts, yielding state-of-the-art performance on three established benchmarks.

The segmentation of the Inferior Alveolar Canal (IAC) plays a central role in maxillofacial surgery, drawing significant attention in the current research. Because of their outstanding results, deep learning methods are widely adopted in the segmentation of 3D medical volumes, including the IAC in Cone Beam Computed Tomography (CBCT) data. One of the main challenges when segmenting large volumes, including those obtained through CBCT scans, arises from the use of patch-based techniques, mandatory to fit memory constraints. Such training approaches compromise neural network performance due to a reduction in the global contextual information. Performance degradation is prominently evident when the target objects are small with respect to the background, as it happens with the inferior alveolar nerve that develops across the mandible, but involves only a few voxels of the entire scan. In order to target this issue and push state-of-the-art performance in the segmentation of the IAC, we propose an innovative approach that exploits spatial information of extracted patches and integrates it into a Transformer architecture. By incorporating prior knowledge about patch location, our model improves state of the art by ~2 points on the Dice score when integrated with the standard U-Net architecture. The source code of our proposal is publicly released.

Despite their remarkable success in medical image segmentation, the life cycle of deep neural networks remains a challenge in clinical applications. These models must be regularly updated to integrate new medical data and customized to meet evolving diagnostic standards, regulatory requirements, commercial needs, and privacy constraints. Model merging offers a promising solution, as it allows working with multiple specialized networks that can be created and combined dynamically instead of relying on monolithic models. While extensively studied in standard 2D classification, the potential of model merging for 3D segmentation remains unexplored. This paper presents an efficient framework that allows effective model merging in the domain of 3D image segmentation. Our approach builds upon theoretical analysis and encourages wide minima during pre-training, which we demonstrate to facilitate subsequent model merging. Using U-Net 3D, we evaluate the method on distinct anatomical structures with the ToothFairy2 and BTCV Abdomen datasets. To support further research, we release the source code and all the model weights in a dedicated repository: https://github.com/LucaLumetti/UNetTransplant

Multi-person pose estimation is the task of detecting and regressing the keypoint coordinates of multiple people in a single image. Significant progress has been achieved in recent years, especially with the introduction of transformer-based end-to-end methods. In this paper, we present DualPose, a novel framework that enhances multi-person pose estimation by leveraging a dual-block transformer decoding architecture. Class prediction and keypoint estimation are split into parallel blocks so each sub-task can be separately improved and the risk of interference is reduced. This architecture improves the precision of keypoint localization and the model's capacity to accurately classify individuals. To improve model performance, the Keypoint-Block uses parallel processing of self-attentions, providing a novel strategy that improves keypoint localization accuracy and precision. Additionally, DualPose incorporates a contrastive denoising (CDN) mechanism, leveraging positive and negative samples to stabilize training and improve robustness. Thanks to CDN, a variety of training samples are created by introducing controlled noise into the ground truth, improving the model's ability to discern between valid and incorrect keypoints. DualPose achieves state-of-the-art results outperforming recent end-to-end methods, as shown by extensive experiments on the MS COCO and CrowdPose datasets. The code and pretrained models are publicly available.

In recent years, several algorithms have been developed for the segmentation of the Inferior Alveolar Canal (IAC) in Cone-Beam Computed Tomography (CBCT) scans. However, the availability of public datasets in this domain is limited, resulting in a lack of comparative evaluation studies on a common benchmark. To address this scientific gap and encourage deep learning research in the field, the ToothFairy challenge was organized within the MICCAI 2023 conference. In this context, a public dataset was released to also serve as a benchmark for future research. The dataset comprises 443 CBCT scans, with voxel-level annotations of the IAC available for 153 of them, making it the largest publicly available dataset of its kind. The participants of the challenge were tasked with developing an algorithm to accurately identify the IAC using the 2D and 3D-annotated scans. This paper presents the details of the challenge and the contributions made by the most promising methods proposed by the participants. It represents the first comprehensive comparative evaluation of IAC segmentation methods on a common benchmark dataset, providing insights into the current state-of-the-art algorithms and outlining future research directions. Furthermore, to ensure reproducibility and promote future developments, an open-source repository that collects the implementations of the best submissions was released.

Model merging has emerged as a crucial technique in Deep Learning, enabling the integration of multiple models into a unified system while preserving performance and scalability. In this respect, the compositional properties of low-rank adaptation techniques (e.g., LoRA) have proven beneficial, as simple averaging LoRA modules yields a single model that mostly integrates the capabilities of all individual modules. Building on LoRA, we take a step further by imposing that the merged model matches the responses of all learned modules. Solving this objective in closed form yields an indeterminate system with A and B as unknown variables, indicating the existence of infinitely many closed-form solutions. To address this challenge, we introduce LoRM, an alternating optimization strategy that trains one LoRA matrix at a time. This allows solving for each unknown variable individually, thus finding a unique solution. We apply our proposed methodology to Federated Class-Incremental Learning (FCIL), ensuring alignment of model responses both between clients and across tasks. Our method demonstrates state-of-the-art performance across a range of FCIL scenarios. The code to reproduce our experiments is available at this http URL.