Tatiana Tommasi

Yawar Siddiqui, Antonio Alliegro, Alexey Artemov et al.

We introduce MeshGPT, a new approach for generating triangle meshes that reflects the compactness typical of artist-created meshes, in contrast to dense triangle meshes extracted by iso-surfacing methods from neural fields. Inspired by recent advances in powerful large language models, we adopt a sequence-based approach to autoregressively generate triangle meshes as sequences of triangles. We first learn a vocabulary of latent quantized embeddings, using graph convolutions, which inform these embeddings of the local mesh geometry and topology. These embeddings are sequenced and decoded into triangles by a decoder, ensuring that they can effectively reconstruct the mesh. A transformer is then trained on this learned vocabulary to predict the index of the next embedding given previous embeddings. Once trained, our model can be autoregressively sampled to generate new triangle meshes, directly generating compact meshes with sharp edges, more closely imitating the efficient triangulation patterns of human-crafted meshes. MeshGPT demonstrates a notable improvement over state of the art mesh generation methods, with a 9% increase in shape coverage and a 30-point enhancement in FID scores across various categories.

Leonardo Iurada, Silvia Bucci, Timothy M. Hospedales et al.

Deep learning-based recognition systems are deployed at scale for several real-world applications that inevitably involve our social life. Although being of great support when making complex decisions, they might capture spurious data correlations and leverage sensitive attributes (e.g. age, gender, ethnicity). How to factor out this information while keeping a high prediction performance is a task with still several open questions, many of which are shared with those of the domain adaptation and generalization literature which focuses on avoiding visual domain biases. In this work, we propose an in-depth study of the relationship between cross-domain learning (CD) and model fairness by introducing a benchmark on face and medical images spanning several demographic groups as well as classification and localization tasks. After having highlighted the limits of the current evaluation metrics, we introduce a new Harmonic Fairness (HF) score to assess jointly how fair and accurate every model is with respect to a reference baseline. Our study covers 14 CD approaches alongside three state-of-the-art fairness algorithms and shows how the former can outperform the latter. Overall, our work paves the way for a more systematic analysis of fairness problems in computer vision. Code available at: https://github.com/iurada/fairness_crossdomain

Gabriele Tiboni, Karol Arndt, Giuseppe Averta et al.

Physics simulators have shown great promise for conveniently learning reinforcement learning policies in safe, unconstrained environments. However, transferring the acquired knowledge to the real world can be challenging due to the reality gap. To this end, several methods have been recently proposed to automatically tune simulator parameters with posterior distributions given real data, for use with domain randomization at training time. These approaches have been shown to work for various robotic tasks under different settings and assumptions. Nevertheless, existing literature lacks a thorough comparison of existing adaptive domain randomization methods with respect to transfer performance and real-data efficiency. In this work, we present an open benchmark for both offline and online methods (SimOpt, BayRn, DROID, DROPO), to shed light on which are most suitable for each setting and task at hand. We found that online methods are limited by the quality of the currently learned policy for the next iteration, while offline methods may sometimes fail when replaying trajectories in simulation with open-loop commands. The code used will be released at https://github.com/gabrieletiboni/adr-benchmark.

Francesco Cappio Borlino, Silvia Bucci, Tatiana Tommasi

The ability to evolve is fundamental for any valuable autonomous agent whose knowledge cannot remain limited to that injected by the manufacturer. Consider for example a home assistant robot: it should be able to incrementally learn new object categories when requested, but also to recognize the same objects in different environments (rooms) and poses (hand-held/on the floor/above furniture), while rejecting unknown ones. Despite its importance, this scenario has started to raise interest in the robotic community only recently and the related research is still in its infancy, with existing experimental testbeds but no tailored methods. With this work, we propose the first learning approach that deals with all the previously mentioned challenges at once by exploiting a single contrastive objective. We show how it learns a feature space perfectly suitable to incrementally include new classes and is able to capture knowledge which generalizes across a variety of visual domains. Our method is endowed with a tailored effective stopping criterion for each learning episode and exploits a self-paced thresholding strategy that provides the classifier with a reliable rejection option. Both these novel contributions are based on the observation of the data statistics and do not need manual tuning. An extensive experimental analysis confirms the effectiveness of the proposed approach in establishing the new state-of-the-art. The code is available at https://github.com/FrancescoCappio/Contrastive_Open_World.

Chiara Plizzari, Gabriele Goletto, Antonino Furnari et al.

What will the future be? We wonder! In this survey, we explore the gap between current research in egocentric vision and the ever-anticipated future, where wearable computing, with outward facing cameras and digital overlays, is expected to be integrated in our every day lives. To understand this gap, the article starts by envisaging the future through character-based stories, showcasing through examples the limitations of current technology. We then provide a mapping between this future and previously defined research tasks. For each task, we survey its seminal works, current state-of-the-art methodologies and available datasets, then reflect on shortcomings that limit its applicability to future research. Note that this survey focuses on software models for egocentric vision, independent of any specific hardware. The paper concludes with recommendations for areas of immediate explorations so as to unlock our path to the future always-on, personalised and life-enhancing egocentric vision.

Antonio Alliegro, Francesco Cappio Borlino, Tatiana Tommasi

In recent years there has been significant progress in the field of 3D learning on classification, detection and segmentation problems. The vast majority of the existing studies focus on canonical closed-set conditions, neglecting the intrinsic open nature of the real-world. This limits the abilities of robots and autonomous systems involved in safety-critical applications that require managing novel and unknown signals. In this context exploiting 3D data can be a valuable asset since it provides rich information about the geometry of perceived objects and scenes. With this paper we provide the first broad study on 3D Open Set learning. We introduce 3DOS: a novel testbed for semantic novelty detection that considers several settings with increasing difficulties in terms of semantic (category) shift, and covers both in-domain (synthetic-to-synthetic, real-to-real) and cross-domain (synthetic-to-real) scenarios. Moreover, we investigate the related 2D Open Set literature to understand if and how its recent improvements are effective on 3D data. Our extensive benchmark positions several algorithms in the same coherent picture, revealing their strengths and limitations. The results of our analysis may serve as a reliable foothold for future tailored 3D Open Set methods.

Gabriele Tiboni, Raffaello Camoriano, Tatiana Tommasi

Popular industrial robotic problems such as spray painting and welding require (i) conditioning on free-shape 3D objects and (ii) planning of multiple trajectories to solve the task. Yet, existing solutions make strong assumptions on the form of input surfaces and the nature of output paths, resulting in limited approaches unable to cope with real-data variability. By leveraging on recent advances in 3D deep learning, we introduce a novel framework capable of dealing with arbitrary 3D surfaces, and handling a variable number of unordered output paths (i.e. unstructured). Our approach predicts local path segments, which can be later concatenated to reconstruct long-horizon paths. We extensively validate the proposed method in the context of robotic spray painting by releasing PaintNet, the first public dataset of expert demonstrations on free-shape 3D objects collected in a real industrial scenario. A thorough experimental analysis demonstrates the capabilities of our model to promptly predict smooth output paths that cover up to 95% of previously unseen object surfaces, even without explicitly optimizing for paint coverage.

Gabriele Tiboni, Andrea Protopapa, Tatiana Tommasi et al.

Soft robots are gaining popularity thanks to their intrinsic safety to contacts and adaptability. However, the potentially infinite number of Degrees of Freedom makes their modeling a daunting task, and in many cases only an approximated description is available. This challenge makes reinforcement learning (RL) based approaches inefficient when deployed on a realistic scenario, due to the large domain gap between models and the real platform. In this work, we demonstrate, for the first time, how Domain Randomization (DR) can solve this problem by enhancing RL policies for soft robots with: i) robustness w.r.t. unknown dynamics parameters; ii) reduced training times by exploiting drastically simpler dynamic models for learning; iii) better environment exploration, which can lead to exploitation of environmental constraints for optimal performance. Moreover, we introduce a novel algorithmic extension to previous adaptive domain randomization methods for the automatic inference of dynamics parameters for deformable objects. We provide an extensive evaluation in simulation on four different tasks and two soft robot designs, opening interesting perspectives for future research on Reinforcement Learning for closed-loop soft robot control.

Niccolò Cavagnero, Fernando Dos Santos, Marco Ciccone et al.

Deep Neural Networks (DNNs) enable a wide series of technological advancements, ranging from clinical imaging, to predictive industrial maintenance and autonomous driving. However, recent findings indicate that transient hardware faults may corrupt the models prediction dramatically. For instance, the radiation-induced misprediction probability can be so high to impede a safe deployment of DNNs models at scale, urging the need for efficient and effective hardening solutions. In this work, we propose to tackle the reliability issue both at training and model design time. First, we show that vanilla models are highly affected by transient faults, that can induce a performances drop up to 37%. Hence, we provide three zero-overhead solutions, based on DNN re-design and re-train, that can improve DNNs reliability to transient faults up to one order of magnitude. We complement our work with extensive ablation studies to quantify the gain in performances of each hardening component.

Gabriele Tiboni, Pascal Klink, Jan Peters et al.

Varying dynamics parameters in simulation is a popular Domain Randomization (DR) approach for overcoming the reality gap in Reinforcement Learning (RL). Nevertheless, DR heavily hinges on the choice of the sampling distribution of the dynamics parameters, since high variability is crucial to regularize the agent's behavior but notoriously leads to overly conservative policies when randomizing excessively. In this paper, we propose a novel approach to address sim-to-real transfer, which automatically shapes dynamics distributions during training in simulation without requiring real-world data. We introduce DOmain RAndomization via Entropy MaximizatiON (DORAEMON), a constrained optimization problem that directly maximizes the entropy of the training distribution while retaining generalization capabilities. In achieving this, DORAEMON gradually increases the diversity of sampled dynamics parameters as long as the probability of success of the current policy is sufficiently high. We empirically validate the consistent benefits of DORAEMON in obtaining highly adaptive and generalizable policies, i.e. solving the task at hand across the widest range of dynamics parameters, as opposed to representative baselines from the DR literature. Notably, we also demonstrate the Sim2Real applicability of DORAEMON through its successful zero-shot transfer in a robotic manipulation setup under unknown real-world parameters.

Francesco Cappio Borlino, Silvia Bucci, Tatiana Tommasi

Semantic novelty detection aims at discovering unknown categories in the test data. This task is particularly relevant in safety-critical applications, such as autonomous driving or healthcare, where it is crucial to recognize unknown objects at deployment time and issue a warning to the user accordingly. Despite the impressive advancements of deep learning research, existing models still need a finetuning stage on the known categories in order to recognize the unknown ones. This could be prohibitive when privacy rules limit data access, or in case of strict memory and computational constraints (e.g. edge computing). We claim that a tailored representation learning strategy may be the right solution for effective and efficient semantic novelty detection. Besides extensively testing state-of-the-art approaches for this task, we propose a novel representation learning paradigm based on relational reasoning. It focuses on learning how to measure semantic similarity rather than recognizing known categories. Our experiments show that this knowledge is directly transferable to a wide range of scenarios, and it can be exploited as a plug-and-play module to convert closed-set recognition models into reliable open-set ones.

Lorenzo Li Lu, Giulia D'Ascenzi, Francesco Cappio Borlino et al.

Standard recognition approaches are unable to deal with novel categories at test time. Their overconfidence on the known classes makes the predictions unreliable for safety-critical applications such as healthcare or autonomous driving. Out-Of-Distribution (OOD) detection methods provide a solution by identifying semantic novelty. Most of these methods leverage a learning stage on the known data, which means training (or fine-tuning) a model to capture the concept of normality. This process is clearly sensitive to the amount of available samples and might be computationally expensive for on-board systems. A viable alternative is that of evaluating similarities in the embedding space produced by large pre-trained models without any further learning effort. We focus exactly on such a fine-tuning-free OOD detection setting. This works presents an in-depth analysis of the recently introduced relational reasoning pre-training and investigates the properties of the learned embedding, highlighting the existence of a correlation between the inter-class feature distance and the OOD detection accuracy. As the class separation depends on the chosen pre-training objective, we propose an alternative loss function to control the inter-class margin, and we show its advantage with thorough experiments.

Leonardo Iurada, Niccolò Cavagnero, Fernando Fernandes Dos Santos et al.

Deep learning models are crucial for autonomous vehicle perception, but their reliability is challenged by algorithmic limitations and hardware faults. We address the latter by examining fault-tolerance in semantic segmentation models. Using established hardware fault models, we evaluate existing hardening techniques both in terms of accuracy and uncertainty and introduce ReLUMax, a novel simple activation function designed to enhance resilience against transient faults. ReLUMax integrates seamlessly into existing architectures without time overhead. Our experiments demonstrate that ReLUMax effectively improves robustness, preserving performance and boosting prediction confidence, thus contributing to the development of reliable autonomous driving systems.

Paolo Rabino, Antonio Alliegro, Francesco Cappio Borlino et al.

Moving deep learning models from the laboratory setting to the open world entails preparing them to handle unforeseen conditions. In several applications the occurrence of novel classes during deployment poses a significant threat, thus it is crucial to effectively detect them. Ideally, this skill should be used when needed without requiring any further computational training effort at every new task. Out-of-distribution detection has attracted significant attention in the last years, however the majority of the studies deal with 2D images ignoring the inherent 3D nature of the real-world and often confusing between domain and semantic novelty. In this work, we focus on the latter, considering the objects geometric structure captured by 3D point clouds regardless of the specific domain. We advance the field by introducing OpenPatch that builds on a large pre-trained model and simply extracts from its intermediate features a set of patch representations that describe each known class. For any new sample, we obtain a novelty score by evaluating whether it can be recomposed mainly by patches of a single known class or rather via the contribution of multiple classes. We present an extensive experimental evaluation of our approach for the task of semantic novelty detection on real-world point cloud samples when the reference known data are synthetic. We demonstrate that OpenPatch excels in both the full and few-shot known sample scenarios, showcasing its robustness across varying pre-training objectives and network backbones. The inherent training-free nature of our method allows for its immediate application to a wide array of real-world tasks, offering a compelling advantage over approaches that need expensive retraining efforts.

Paolo Rabino, Tatiana Tommasi

Interacting with real-world cluttered scenes pose several challenges to robotic agents that need to understand complex spatial dependencies among the observed objects to determine optimal pick sequences or efficient object retrieval strategies. Existing solutions typically manage simplified scenarios and focus on predicting pairwise object relationships following an initial object detection phase, but often overlook the global context or struggle with handling redundant and missing object relations. In this work, we present a modern take on visual relational reasoning for grasp planning. We introduce D3GD, a novel testbed that includes bin picking scenes with up to 35 objects from 97 distinct categories. Additionally, we propose D3G, a new end-to-end transformer-based dependency graph generation model that simultaneously detects objects and produces an adjacency matrix representing their spatial relationships. Recognizing the limitations of standard metrics, we employ the Average Precision of Relationships for the first time to evaluate model performance, conducting an extensive experimental benchmark. The obtained results establish our approach as the new state-of-the-art for this task, laying the foundation for future research in robotic manipulation. We publicly release the code and dataset at https://paolotron.github.io/d3g.github.io.

Leonardo Iurada, Marco Ciccone, Tatiana Tommasi

Recent advances in neural network pruning have shown how it is possible to reduce the computational costs and memory demands of deep learning models before training. We focus on this framework and propose a new pruning at initialization algorithm that leverages the Neural Tangent Kernel (NTK) theory to align the training dynamics of the sparse network with that of the dense one. Specifically, we show how the usually neglected data-dependent component in the NTK's spectrum can be taken into account by providing an analytical upper bound to the NTK's trace obtained by decomposing neural networks into individual paths. This leads to our Path eXclusion (PX), a foresight pruning method designed to preserve the parameters that mostly influence the NTK's trace. PX is able to find lottery tickets (i.e. good paths) even at high sparsity levels and largely reduces the need for additional training. When applied to pre-trained models it extracts subnetworks directly usable for several downstream tasks, resulting in performance comparable to those of the dense counterpart but with substantial cost and computational savings. Code available at: https://github.com/iurada/px-ntk-pruning

Silvia Bucci, Francesco Cappio Borlino, Barbara Caputo et al.

Vision systems trained in closed-world scenarios fail when presented with new environmental conditions, new data distributions, and novel classes at deployment time. How to move towards open-world learning is a long-standing research question. The existing solutions mainly focus on specific aspects of the problem (single domain Open-Set, multi-domain Closed-Set), or propose complex strategies which combine several losses and manually tuned hyperparameters. In this work, we tackle multi-source Open-Set domain adaptation by introducing HyMOS: a straightforward model that exploits the power of contrastive learning and the properties of its hyperspherical feature space to correctly predict known labels on the target, while rejecting samples belonging to any unknown class. HyMOS includes style transfer among the instance transformations of contrastive learning to get domain invariance while avoiding the risk of negative-transfer. A self-paced threshold is defined on the basis of the observed data distribution and updates online during training, allowing to handle the known-unknown separation. We validate our method over three challenging datasets. The obtained results show that HyMOS outperforms several competitors, defining the new state-of-the-art. Our code is available at https://github.com/silvia1993/HyMOS.

Paolo Rabino, Gabriele Tiboni, Tatiana Tommasi

Object-Centric Motion Generation (OCMG) is instrumental in advancing automated manufacturing processes, particularly in domains requiring high-precision expert robotic motions, such as spray painting and welding. To realize effective automation, robust algorithms are essential for generating extended, object-aware trajectories across intricate 3D geometries. However, contemporary OCMG techniques are either based on ad-hoc heuristics or employ learning-based pipelines that are still reliant on sensitive post-processing steps to generate executable paths. We introduce FoldPath, a novel, end-to-end, neural field based method for OCMG. Unlike prior deep learning approaches that predict discrete sequences of end-effector waypoints, FoldPath learns the robot motion as a continuous function, thus implicitly encoding smooth output paths. This paradigm shift eliminates the need for brittle post-processing steps that concatenate and order the predicted discrete waypoints. Particularly, our approach demonstrates superior predictive performance compared to recently proposed learning-based methods, and attains generalization capabilities even in real industrial settings, where only a limited amount of 70 expert samples are provided. We validate FoldPath through comprehensive experiments in a realistic simulation environment and introduce new, rigorous metrics designed to comprehensively evaluate long-horizon robotic paths, thus advancing the OCMG task towards practical maturity.

Antonio Alliegro, Yawar Siddiqui, Tatiana Tommasi et al.

We introduce PolyDiff, the first diffusion-based approach capable of directly generating realistic and diverse 3D polygonal meshes. In contrast to methods that use alternate 3D shape representations (e.g. implicit representations), our approach is a discrete denoising diffusion probabilistic model that operates natively on the polygonal mesh data structure. This enables learning of both the geometric properties of vertices and the topological characteristics of faces. Specifically, we treat meshes as quantized triangle soups, progressively corrupted with categorical noise in the forward diffusion phase. In the reverse diffusion phase, a transformer-based denoising network is trained to revert the noising process, restoring the original mesh structure. At inference, new meshes can be generated by applying this denoising network iteratively, starting with a completely noisy triangle soup. Consequently, our model is capable of producing high-quality 3D polygonal meshes, ready for integration into downstream 3D workflows. Our extensive experimental analysis shows that PolyDiff achieves a significant advantage (avg. FID and JSD improvement of 18.2 and 5.8 respectively) over current state-of-the-art methods.

Stephany Ortuno-Chanelo, Paolo Rabino, Enrico Civitelli et al.

Vision-based models for robotic grasping automate critical, repetitive, and draining industrial tasks. Existing approaches are typically limited in two ways: they either target a single gripper and are potentially applied on costly dual-arm setups, or rely on custom hybrid grippers that require ad-hoc learning procedures with logic that cannot be transferred across tasks, restricting their general applicability. In this work, we present MultiGraspNet, a novel multitask 3D deep learning method that predicts feasible poses simultaneously for parallel and vacuum grippers within a unified framework, enabling a single robot to handle multiple end effectors. The model is trained on the richly annotated GraspNet-1Billion and SuctionNet-1Billion datasets, which have been aligned for the purpose, and generates graspability masks quantifying the suitability of each scene point for successful grasps. By sharing early-stage features while maintaining gripper-specific refiners, MultiGraspNet effectively leverages complementary information across grasping modalities, enhancing robustness and adaptability in cluttered scenes. We characterize MultiGraspNet's performance with an extensive experimental analysis, demonstrating its competitiveness with single-task models on relevant benchmarks. We run real-world experiments on a single-arm multi-gripper robotic setup showing that our approach outperforms the vacuum baseline, grasping 16% percent more seen objects and 32% more of the novel ones, while obtaining competitive results for the parallel task.

Leonardo Iurada, Marco Ciccone, Tatiana Tommasi

Task arithmetic has emerged as a promising approach for editing models by representing task-specific knowledge as composable task vectors. However, existing methods rely on network linearization to derive task vectors, leading to computational bottlenecks during training and inference. Moreover, linearization alone does not ensure weight disentanglement, the key property that enables conflict-free composition of task vectors. To address this, we propose TaLoS which allows to build sparse task vectors with minimal interference without requiring explicit linearization and sharing information across tasks. We find that pre-trained models contain a subset of parameters with consistently low gradient sensitivity across tasks, and that sparsely updating only these parameters allows for promoting weight disentanglement during fine-tuning. Our experiments prove that TaLoS improves training and inference efficiency while outperforming current methods in task addition and negation. By enabling modular parameter editing, our approach fosters practical deployment of adaptable foundation models in real-world applications.

Simone Alberto Peirone, Francesca Pistilli, Antonio Alliegro et al.

Our comprehension of video streams depicting human activities is naturally multifaceted: in just a few moments, we can grasp what is happening, identify the relevance and interactions of objects in the scene, and forecast what will happen soon, everything all at once. To endow autonomous systems with such a holistic perception, learning how to correlate concepts, abstract knowledge across diverse tasks, and leverage tasks synergies when learning novel skills is essential. A significant step in this direction is EgoPack, a unified framework for understanding human activities across diverse tasks with minimal overhead. EgoPack promotes information sharing and collaboration among downstream tasks, essential for efficiently learning new skills. In this paper, we introduce Hier-EgoPack, which advances EgoPack by enabling reasoning also across diverse temporal granularities, which expands its applicability to a broader range of downstream tasks. To achieve this, we propose a novel hierarchical architecture for temporal reasoning equipped with a GNN layer specifically designed to tackle the challenges of multi-granularity reasoning effectively. We evaluate our approach on multiple Ego4d benchmarks involving both clip-level and frame-level reasoning, demonstrating how our hierarchical unified architecture effectively solves these diverse tasks simultaneously.

Gabriele Tiboni, Raffaello Camoriano, Tatiana Tommasi

Object-Centric Motion Generation (OCMG) plays a key role in a variety of industrial applications$\unicode{x2014}$such as robotic spray painting and welding$\unicode{x2014}$requiring efficient, scalable, and generalizable algorithms to plan multiple long-horizon trajectories over free-form 3D objects. However, existing solutions rely on specialized heuristics, expensive optimization routines, or restrictive geometry assumptions that limit their adaptability to real-world scenarios. In this work, we introduce a novel, fully data-driven framework that tackles OCMG directly from 3D point clouds, learning to generalize expert path patterns across free-form surfaces. We propose MaskPlanner, a deep learning method that predicts local path segments for a given object while simultaneously inferring "path masks" to group these segments into distinct paths. This design induces the network to capture both local geometric patterns and global task requirements in a single forward pass. Extensive experimentation on a realistic robotic spray painting scenario shows that our approach attains near-complete coverage (above 99%) for unseen objects, while it remains task-agnostic and does not explicitly optimize for paint deposition. Moreover, our real-world validation on a 6-DoF specialized painting robot demonstrates that the generated trajectories are directly executable and yield expert-level painting quality. Our findings crucially highlight the potential of the proposed learning method for OCMG to reduce engineering overhead and seamlessly adapt to several industrial use cases.

Qitian Ma, Shyam Nanda Rai, Carlo Masone et al.

In the domain of computer vision, semantic segmentation emerges as a fundamental application within machine learning, wherein individual pixels of an image are classified into distinct semantic categories. This task transcends traditional accuracy metrics by incorporating uncertainty quantification, a critical measure for assessing the reliability of each segmentation prediction. Such quantification is instrumental in facilitating informed decision-making, particularly in applications where precision is paramount. Within this nuanced framework, the metric known as PAvPU (Patch Accuracy versus Patch Uncertainty) has been developed as a specialized tool for evaluating entropy-based uncertainty in image segmentation tasks. However, our investigation identifies three core deficiencies within the PAvPU framework and proposes robust solutions aimed at refining the metric. By addressing these issues, we aim to enhance the reliability and applicability of uncertainty quantification, especially in scenarios that demand high levels of safety and accuracy, thus contributing to the advancement of semantic segmentation methodologies in critical applications.

Silvio Chito, Paolo Rabino, Tatiana Tommasi

The recently introduced odd-one-out anomaly detection task involves identifying the odd-looking instances within a multi-object scene. This problem presents several challenges for modern deep learning models, demanding spatial reasoning across multiple views and relational reasoning to understand context and generalize across varying object categories and layouts. We argue that these challenges must be addressed with efficiency in mind. To this end, we propose a DINO-based model that reduces the number of parameters by one third and shortens training time by a factor of three compared to the current state-of-the-art, while maintaining competitive performance. Our experimental evaluation also introduces a Multimodal Large Language Model baseline, providing insights into its current limitations in structured visual reasoning tasks. The project page can be found at https://silviochito.github.io/EfficientOddOneOut/

Leonardo Iurada, Beatrice Occhiena, Tatiana Tommasi

The widespread availability of pre-trained vision models has enabled numerous deep learning applications through their transferable representations. However, their computational and storage costs often limit practical deployment. Pruning-at-Initialization has emerged as a promising approach to compress models before training, enabling efficient task-specific adaptation. While conventional wisdom suggests that effective pruning requires task-specific data, this creates a challenge when downstream tasks are unknown in advance. In this paper, we investigate how data influences the pruning of pre-trained vision models. Surprisingly, pruning on one task retains the model's zero-shot performance also on unseen tasks. Furthermore, fine-tuning these pruned models not only improves performance on original seen tasks but can recover held-out tasks' performance. We attribute this phenomenon to the favorable loss landscapes induced by extensive pre-training on large-scale datasets.

Leonardo Iurada, Simone Bombari, Tatiana Tommasi et al.

Large-scale deep learning models are known to memorize parts of the training set. In machine learning theory, memorization is often framed as interpolation or label fitting, and classical results show that this can be achieved when the number of parameters $p$ in the model is larger than the number of training samples $n$. In this work, we consider memorization from the perspective of data reconstruction, demonstrating that this can be achieved when $p$ is larger than $dn$, where $d$ is the dimensionality of the data. More specifically, we show that, in the random features model, when $p \gg dn$, the subspace spanned by the training samples in feature space gives sufficient information to identify the individual samples in input space. Our analysis suggests an optimization method to reconstruct the dataset from the model parameters, and we demonstrate that this method performs well on various architectures (random features, two-layer fully-connected and deep residual networks). Our results reveal a law of data reconstruction, according to which the entire training dataset can be recovered as $p$ exceeds the threshold $dn$.

Simone Alberto Peirone, Francesca Pistilli, Antonio Alliegro et al.

Our comprehension of video streams depicting human activities is naturally multifaceted: in just a few moments, we can grasp what is happening, identify the relevance and interactions of objects in the scene, and forecast what will happen soon, everything all at once. To endow autonomous systems with such holistic perception, learning how to correlate concepts, abstract knowledge across diverse tasks, and leverage tasks synergies when learning novel skills is essential. In this paper, we introduce Hier-EgoPack, a unified framework able to create a collection of task perspectives that can be carried across downstream tasks and used as a potential source of additional insights, as a backpack of skills that a robot can carry around and use when needed.

Antonio Alliegro, Francesca Pistilli, Tatiana Tommasi et al.

Forecasting human-environment interactions in daily activities is challenging due to the high variability of human behavior. While predicting directly from videos is possible, it is limited by confounding factors like irrelevant objects or background noise that do not contribute to the interaction. A promising alternative is using Scene Graphs (SGs) to track only the relevant elements. However, current methods for forecasting future SGs face significant challenges and often rely on unrealistic assumptions, such as fixed objects over time, limiting their applicability to long-term activities where interacted objects may appear or disappear. In this paper, we introduce FORESCENE, a novel framework for Scene Graph Anticipation (SGA) that predicts both object and relationship evolution over time. FORESCENE encodes observed video segments into a latent representation using a tailored Graph Auto-Encoder and forecasts future SGs using a Latent Diffusion Model (LDM). Our approach enables continuous prediction of interaction dynamics without making assumptions on the graph's content or structure. We evaluate FORESCENE on the Action Genome dataset, where it outperforms existing SGA methods while solving a significantly more complex task.

Tatiana Tommasi, Silvia Bucci, Barbara Caputo et al.

Thanks to the great progress of machine learning in the last years, several Artificial Intelligence (AI) techniques have been increasingly moving from the controlled research laboratory settings to our everyday life. AI is clearly supportive in many decision-making scenarios, but when it comes to sensitive areas such as health care, hiring policies, education, banking or justice, with major impact on individuals and society, it becomes crucial to establish guidelines on how to design, develop, deploy and monitor this technology. Indeed the decision rules elaborated by machine learning models are data-driven and there are multiple ways in which discriminatory biases can seep into data. Algorithms trained on those data incur the risk of amplifying prejudices and societal stereotypes by over associating protected attributes such as gender, ethnicity or disabilities with the prediction task. Starting from the extensive experience of the National Metrology Institute on measurement standards and certification roadmaps, and of Politecnico di Torino on machine learning as well as methods for domain bias evaluation and mastering, we propose a first joint effort to define the operational steps needed for AI fairness certification. Specifically we will overview the criteria that should be met by an AI system before coming into official service and the conformity assessment procedures useful to monitor its functioning for fair decisions.

Antonio Alliegro, Diego Valsesia, Giulia Fracastoro et al.

In this paper, we present a deep learning model that exploits the power of self-supervision to perform 3D point cloud completion, estimating the missing part and a context region around it. Local and global information are encoded in a combined embedding. A denoising pretext task provides the network with the needed local cues, decoupled from the high-level semantics and naturally shared over multiple classes. On the other hand, contrastive learning maximizes the agreement between variants of the same shape with different missing portions, thus producing a representation which captures the global appearance of the shape. The combined embedding inherits category-agnostic properties from the chosen pretext tasks. Differently from existing approaches, this allows to better generalize the completion properties to new categories unseen at training time. Moreover, while decoding the obtained joint representation, we better blend the reconstructed missing part with the partial shape by paying attention to its known surrounding region and reconstructing this frame as auxiliary objective. Our extensive experiments and detailed ablation on the ShapeNet dataset show the effectiveness of each part of the method with new state of the art results. Our quantitative and qualitative analysis confirms how our approach is able to work on novel categories without relying neither on classification and shape symmetry priors, nor on adversarial training procedures.

Andrea Ferreri, Silvia Bucci, Tatiana Tommasi

Scene recognition is one of the basic problems in computer vision research with extensive applications in robotics. When available, depth images provide helpful geometric cues that complement the RGB texture information and help to identify discriminative scene image features. Depth sensing technology developed fast in the last years and a great variety of 3D cameras have been introduced, each with different acquisition properties. However, those properties are often neglected when targeting big data collections, so multi-modal images are gathered disregarding their original nature. In this work, we put under the spotlight the existence of a possibly severe domain shift issue within multi-modality scene recognition datasets. As a consequence, a scene classification model trained on one camera may not generalize on data from a different camera, only providing a low recognition performance. Starting from the well-known SUN RGB-D dataset, we designed an experimental testbed to study this problem and we use it to benchmark the performance of existing methods. Finally, we introduce a novel adaptive scene recognition approach that leverages self-supervised translation between modalities. Indeed, learning to go from RGB to depth and vice-versa is an unsupervised procedure that can be trained jointly on data of multiple cameras and may help to bridge the gap among the extracted feature distributions. Our experimental results confirm the effectiveness of the proposed approach.

Francesco Cappio Borlino, Antonio D'Innocente, Tatiana Tommasi

Despite being very powerful in standard learning settings, deep learning models can be extremely brittle when deployed in scenarios different from those on which they were trained. Domain generalization methods investigate this problem and data augmentation strategies have shown to be helpful tools to increase data variability, supporting model robustness across domains. In our work we focus on style transfer data augmentation and we present how it can be implemented with a simple and inexpensive strategy to improve generalization. Moreover, we analyze the behavior of current state of the art domain generalization methods when integrated with this augmentation solution: our thorough experimental evaluation shows that their original effect almost always disappears with respect to the augmented baseline. This issue open new scenarios for domain generalization research, highlighting the need of novel methods properly able to take advantage of the introduced data variability.

Silvia Bucci, Antonio D'Innocente, Yujun Liao et al.

Human adaptability relies crucially on learning and merging knowledge from both supervised and unsupervised tasks: the parents point out few important concepts, but then the children fill in the gaps on their own. This is particularly effective, because supervised learning can never be exhaustive and thus learning autonomously allows to discover invariances and regularities that help to generalize. In this paper we propose to apply a similar approach to the problem of object recognition across domains: our model learns the semantic labels in a supervised fashion, and broadens its understanding of the data by learning from self-supervised signals on the same images. This secondary task helps the network to focus on object shapes, learning concepts like spatial orientation and part correlation, while acting as a regularizer for the classification task over multiple visual domains. Extensive experiments confirm our intuition and show that our multi-task method combining supervised and self-supervised knowledge shows competitive results with respect to more complex domain generalization and adaptation solutions. It also proves its potential in the novel and challenging predictive and partial domain adaptation scenarios.

Silvia Bucci, Mohammad Reza Loghmani, Tatiana Tommasi

Open Set Domain Adaptation (OSDA) bridges the domain gap between a labeled source domain and an unlabeled target domain, while also rejecting target classes that are not present in the source. To avoid negative transfer, OSDA can be tackled by first separating the known/unknown target samples and then aligning known target samples with the source data. We propose a novel method to addresses both these problems using the self-supervised task of rotation recognition. Moreover, we assess the performance with a new open set metric that properly balances the contribution of recognizing the known classes and rejecting the unknown samples. Comparative experiments with existing OSDA methods on the standard Office-31 and Office-Home benchmarks show that: (i) our method outperforms its competitors, (ii) reproducibility for this field is a crucial issue to tackle, (iii) our metric provides a reliable tool to allow fair open set evaluation.

Antonio D'Innocente, Francesco Cappio Borlino, Silvia Bucci et al.

Despite impressive progress in object detection over the last years, it is still an open challenge to reliably detect objects across visual domains. Although the topic has attracted attention recently, current approaches all rely on the ability to access a sizable amount of target data for use at training time. This is a heavy assumption, as often it is not possible to anticipate the domain where a detector will be used, nor to access it in advance for data acquisition. Consider for instance the task of monitoring image feeds from social media: as every image is created and uploaded by a different user it belongs to a different target domain that is impossible to foresee during training. This paper addresses this setting, presenting an object detection algorithm able to perform unsupervised adaption across domains by using only one target sample, seen at test time. We achieve this by introducing a multi-task architecture that one-shot adapts to any incoming sample by iteratively solving a self-supervised task on it. We further enhance this auxiliary adaptation with cross-task pseudo-labeling. A thorough benchmark analysis against the most recent cross-domain detection methods and a detailed ablation study show the advantage of our method, which sets the state-of-the-art in the defined one-shot scenario.

Lia Morra, Luca Piano, Fabrizio Lamberti et al.

Deep learning has thrived by training on large-scale datasets. However, in many applications, as for medical image diagnosis, getting massive amount of data is still prohibitive due to privacy, lack of acquisition homogeneity and annotation cost. In this scenario, transfer learning from natural image collections is a standard practice that attempts to tackle shape, texture and color discrepancies all at once through pretrained model fine-tuning. In this work, we propose to disentangle those challenges and design a dedicated network module that focuses on color adaptation. We combine learning from scratch of the color module with transfer learning of different classification backbones, obtaining an end-to-end, easy-to-train architecture for diagnostic image recognition on X-ray images. Extensive experiments showed how our approach is particularly efficient in case of data scarcity and provides a new path for further transferring the learned color information across multiple medical datasets.

Antonio Alliegro, Davide Boscaini, Tatiana Tommasi

Point cloud processing and 3D shape understanding are very challenging tasks for which deep learning techniques have demonstrated great potentials. Still further progresses are essential to allow artificial intelligent agents to interact with the real world, where the amount of annotated data may be limited and integrating new sources of knowledge becomes crucial to support autonomous learning. Here we consider several possible scenarios involving synthetic and real-world point clouds where supervised learning fails due to data scarcity and large domain gaps. We propose to enrich standard feature representations by leveraging self-supervision through a multi-task model that can solve a 3D puzzle while learning the main task of shape classification or part segmentation. An extensive analysis investigating few-shot, transfer learning and cross-domain settings shows the effectiveness of our approach with state-of-the-art results for 3D shape classification and part segmentation.

Antonio D'Innocente, Silvia Bucci, Barbara Caputo et al.

Although deep networks have significantly increased the performance of visual recognition methods, it is still challenging to achieve the robustness across visual domains that is necessary for real-world applications. To tackle this issue, research on domain adaptation and generalization has flourished over the last decade. An important aspect to consider when assessing the work done in the literature so far is the amount of data annotation necessary for training each approach, both at the source and target level. In this paper we argue that the data annotation overload should be minimal, as it is costly. Hence, we propose to use self-supervised learning to achieve domain generalization and adaptation. We consider learning regularities from non annotated data as an auxiliary task, and cast the problem within an Auxiliary Learning principled framework. Moreover, we suggest to further exploit the ability to learn about visual domains from non annotated images by learning from target data while testing, as data are presented to the algorithm one sample at a time. Results on three different scenarios confirm the value of our approach.

Silvia Bucci, Antonio D'Innocente, Tatiana Tommasi

Domain adaptation approaches have shown promising results in reducing the marginal distribution difference among visual domains. They allow to train reliable models that work over datasets of different nature (photos, paintings etc), but they still struggle when the domains do not share an identical label space. In the partial domain adaptation setting, where the target covers only a subset of the source classes, it is challenging to reduce the domain gap without incurring in negative transfer. Many solutions just keep the standard domain adaptation techniques by adding heuristic sample weighting strategies. In this work we show how the self-supervisory signal obtained from the spatial co-location of patches can be used to define a side task that supports adaptation regardless of the exact label sharing condition across domains. We build over a recent work that introduced a jigsaw puzzle task for domain generalization: we describe how to reformulate this approach for partial domain adaptation and we show how it boosts existing adaptive solutions when combined with them. The obtained experimental results on three datasets supports the effectiveness of our approach.

Fabio Maria Carlucci, Antonio D'Innocente, Silvia Bucci et al.

Human adaptability relies crucially on the ability to learn and merge knowledge both from supervised and unsupervised learning: the parents point out few important concepts, but then the children fill in the gaps on their own. This is particularly effective, because supervised learning can never be exhaustive and thus learning autonomously allows to discover invariances and regularities that help to generalize. In this paper we propose to apply a similar approach to the task of object recognition across domains: our model learns the semantic labels in a supervised fashion, and broadens its understanding of the data by learning from self-supervised signals how to solve a jigsaw puzzle on the same images. This secondary task helps the network to learn the concepts of spatial correlation while acting as a regularizer for the classification task. Multiple experiments on the PACS, VLCS, Office-Home and digits datasets confirm our intuition and show that this simple method outperforms previous domain generalization and adaptation solutions. An ablation study further illustrates the inner workings of our approach.

Fabio M. Carlucci, Paolo Russo, Tatiana Tommasi et al.

The ability to generalize across visual domains is crucial for the robustness of artificial recognition systems. Although many training sources may be available in real contexts, the access to even unlabeled target samples cannot be taken for granted, which makes standard unsupervised domain adaptation methods inapplicable in the wild. In this work we investigate how to exploit multiple sources by hallucinating a deep visual domain composed of images, possibly unrealistic, able to maintain categorical knowledge while discarding specific source styles. The produced agnostic images are the result of a deep architecture that applies pixel adaptation on the original source data guided by two adversarial domain classifier branches at image and feature level. Our approach is conceived to learn only from source data, but it seamlessly extends to the use of unlabeled target samples. Remarkable results for both multi-source domain adaptation and domain generalization support the power of hallucinating agnostic images in this framework.

Gabriele Angeletti, Barbara Caputo, Tatiana Tommasi

A commercial robot, trained by its manufacturer to recognize a predefined number and type of objects, might be used in many settings, that will in general differ in their illumination conditions, background, type and degree of clutter, and so on. Recent computer vision works tackle this generalization issue through domain adaptation methods, assuming as source the visual domain where the system is trained and as target the domain of deployment. All approaches assume to have access to images from all classes of the target during training, an unrealistic condition in robotics applications. We address this issue proposing an algorithm that takes into account the specific needs of robot vision. Our intuition is that the nature of the domain shift experienced mostly in robotics is local. We exploit this through the learning of maps that spatially ground the domain and quantify the degree of shift, embedded into an end-to-end deep domain adaptation architecture. By explicitly localizing the roots of the domain shift we significantly reduce the number of parameters of the architecture to tune, we gain the flexibility necessary to deal with subset of categories in the target domain at training time, and we provide a clear feedback on the rationale behind any classification decision, which can be exploited in human-robot interactions. Experiments on two different settings of the iCub World database confirm the suitability of our method for robot vision.

Paolo Russo, Fabio Maria Carlucci, Tatiana Tommasi et al.

The effectiveness of generative adversarial approaches in producing images according to a specific style or visual domain has recently opened new directions to solve the unsupervised domain adaptation problem. It has been shown that source labeled images can be modified to mimic target samples making it possible to train directly a classifier in the target domain, despite the original lack of annotated data. Inverse mappings from the target to the source domain have also been evaluated but only passing through adapted feature spaces, thus without new image generation. In this paper we propose to better exploit the potential of generative adversarial networks for adaptation by introducing a novel symmetric mapping among domains. We jointly optimize bi-directional image transformations combining them with target self-labeling. Moreover we define a new class consistency loss that aligns the generators in the two directions imposing to conserve the class identity of an image passing through both domain mappings. A detailed qualitative and quantitative analysis of the reconstructed images confirm the power of our approach. By integrating the two domain specific classifiers obtained with our bi-directional network we exceed previous state-of-the-art unsupervised adaptation results on four different benchmark datasets.

Francesco Orabona, Tatiana Tommasi

Deep learning methods achieve state-of-the-art performance in many application scenarios. Yet, these methods require a significant amount of hyperparameters tuning in order to achieve the best results. In particular, tuning the learning rates in the stochastic optimization process is still one of the main bottlenecks. In this paper, we propose a new stochastic gradient descent procedure for deep networks that does not require any learning rate setting. Contrary to previous methods, we do not adapt the learning rates nor we make use of the assumed curvature of the objective function. Instead, we reduce the optimization process to a game of betting on a coin and propose a learning-rate-free optimal algorithm for this scenario. Theoretical convergence is proven for convex and quasi-convex functions and empirical evidence shows the advantage of our algorithm over popular stochastic gradient algorithms.

Nizar Massouh, Francesca Babiloni, Tatiana Tommasi et al.

Deep networks thrive when trained on large scale data collections. This has given ImageNet a central role in the development of deep architectures for visual object classification. However, ImageNet was created during a specific period in time, and as such it is prone to aging, as well as dataset bias issues. Moving beyond fixed training datasets will lead to more robust visual systems, especially when deployed on robots in new environments which must train on the objects they encounter there. To make this possible, it is important to break free from the need for manual annotators. Recent work has begun to investigate how to use the massive amount of images available on the Web in place of manual image annotations. We contribute to this research thread with two findings: (1) a study correlating a given level of noisily labels to the expected drop in accuracy, for two deep architectures, on two different types of noise, that clearly identifies GoogLeNet as a suitable architecture for learning from Web data; (2) a recipe for the creation of Web datasets with minimal noise and maximum visual variability, based on a visual and natural language processing concept expansion strategy. By combining these two results, we obtain a method for learning powerful deep object models automatically from the Web. We confirm the effectiveness of our approach through object categorization experiments using our Web-derived version of ImageNet on a popular robot vision benchmark database, and on a lifelong object discovery task on a mobile robot.

Tatiana Tommasi, Arun Mallya, Bryan Plummer et al.

This paper presents an approach for answering fill-in-the-blank multiple choice questions from the Visual Madlibs dataset. Instead of generic and commonly used representations trained on the ImageNet classification task, our approach employs a combination of networks trained for specialized tasks such as scene recognition, person activity classification, and attribute prediction. We also present a method for localizing phrases from candidate answers in order to provide spatial support for feature extraction. We map each of these features, together with candidate answers, to a joint embedding space through normalized canonical correlation analysis (nCCA). Finally, we solve an optimization problem to learn to combine scores from nCCA models trained on multiple cues to select the best answer. Extensive experimental results show a significant improvement over the previous state of the art and confirm that answering questions from a wide range of types benefits from examining a variety of image cues and carefully choosing the spatial support for feature extraction.

Tatiana Tommasi, Arun Mallya, Bryan Plummer et al.

This paper focuses on answering fill-in-the-blank style multiple choice questions from the Visual Madlibs dataset. Previous approaches to Visual Question Answering (VQA) have mainly used generic image features from networks trained on the ImageNet dataset, despite the wide scope of questions. In contrast, our approach employs features derived from networks trained for specialized tasks of scene classification, person activity prediction, and person and object attribute prediction. We also present a method for selecting sub-regions of an image that are relevant for evaluating the appropriateness of a putative answer. Visual features are computed both from the whole image and from local regions, while sentences are mapped to a common space using a simple normalized canonical correlation analysis (CCA) model. Our results show a significant improvement over the previous state of the art, and indicate that answering different question types benefits from examining a variety of image cues and carefully choosing informative image sub-regions.

Tatiana Tommasi, Martina Lanzi, Paolo Russo et al.

In this paper we focus on the spatial nature of visual domain shift, attempting to learn where domain adaptation originates in each given image of the source and target set. We borrow concepts and techniques from the CNN visualization literature, and learn domainnes maps able to localize the degree of domain specificity in images. We derive from these maps features related to different domainnes levels, and we show that by considering them as a preprocessing step for a domain adaptation algorithm, the final classification performance is strongly improved. Combined with the whole image representation, these features provide state of the art results on the Office dataset.

Efstratios Gavves, Thomas Mensink, Tatiana Tommasi et al.

How can we reuse existing knowledge, in the form of available datasets, when solving a new and apparently unrelated target task from a set of unlabeled data? In this work we make a first contribution to answer this question in the context of image classification. We frame this quest as an active learning problem and use zero-shot classifiers to guide the learning process by linking the new task to the existing classifiers. By revisiting the dual formulation of adaptive SVM, we reveal two basic conditions to choose greedily only the most relevant samples to be annotated. On this basis we propose an effective active learning algorithm which learns the best possible target classification model with minimum human labeling effort. Extensive experiments on two challenging datasets show the value of our approach compared to the state-of-the-art active learning methodologies, as well as its potential to reuse past datasets with minimal effort for future tasks.