Andrea Pilzer

Xuanlong Yu, Yi Zuo, Zitao Wang et al. · cmu

This paper outlines the winning solutions employed in addressing the MUAD uncertainty quantification challenge held at ICCV 2023. The challenge was centered around semantic segmentation in urban environments, with a particular focus on natural adversarial scenarios. The report presents the results of 19 submitted entries, with numerous techniques drawing inspiration from cutting-edge uncertainty quantification methodologies presented at prominent conferences in the fields of computer vision and machine learning and journals over the past few years. Within this document, the challenge is introduced, shedding light on its purpose and objectives, which primarily revolved around enhancing the robustness of semantic segmentation in urban scenes under varying natural adversarial conditions. The report then delves into the top-performing solutions. Moreover, the document aims to provide a comprehensive overview of the diverse solutions deployed by all participants. By doing so, it seeks to offer readers a deeper insight into the array of strategies that can be leveraged to effectively handle the inherent uncertainties associated with autonomous driving and semantic segmentation, especially within urban environments.

Lorenzo Baraldi, Roberto Amoroso, Marcella Cornia et al.

The use of self-supervised pre-training has emerged as a promising approach to enhance the performance of many different visual tasks. In this context, recent approaches have employed the Masked Image Modeling paradigm, which pre-trains a backbone by reconstructing visual tokens associated with randomly masked image patches. This masking approach, however, introduces noise into the input data during pre-training, leading to discrepancies that can impair performance during the fine-tuning phase. Furthermore, input masking neglects the dependencies between corrupted patches, increasing the inconsistencies observed in downstream fine-tuning tasks. To overcome these issues, we propose a new self-supervised pre-training approach, named Masked and Permuted Vision Transformer (MaPeT), that employs autoregressive and permuted predictions to capture intra-patch dependencies. In addition, MaPeT employs auxiliary positional information to reduce the disparity between the pre-training and fine-tuning phases. In our experiments, we employ a fair setting to ensure reliable and meaningful comparisons and conduct investigations on multiple visual tokenizers, including our proposed $k$-CLIP which directly employs discretized CLIP features. Our results demonstrate that MaPeT achieves competitive performance on ImageNet, compared to baselines and competitors under the same model setting. We release an implementation of our code and models at https://github.com/aimagelab/MaPeT.

Subhankar Roy, Martin Trapp, Andrea Pilzer et al.

Source-free domain adaptation (SFDA) aims to adapt a classifier to an unlabelled target data set by only using a pre-trained source model. However, the absence of the source data and the domain shift makes the predictions on the target data unreliable. We propose quantifying the uncertainty in the source model predictions and utilizing it to guide the target adaptation. For this, we construct a probabilistic source model by incorporating priors on the network parameters inducing a distribution over the model predictions. Uncertainties are estimated by employing a Laplace approximation and incorporated to identify target data points that do not lie in the source manifold and to down-weight them when maximizing the mutual information on the target data. Unlike recent works, our probabilistic treatment is computationally lightweight, decouples source training and target adaptation, and requires no specialized source training or changes of the model architecture. We show the advantages of uncertainty-guided SFDA over traditional SFDA in the closed-set and open-set settings and provide empirical evidence that our approach is more robust to strong domain shifts even without tuning.

Sara Papi, Marco Gaido, Andrea Pilzer et al.

Despite its crucial role in research experiments, code correctness is often presumed only on the basis of the perceived quality of results. This assumption comes with the risk of erroneous outcomes and potentially misleading findings. To address this issue, we posit that the current focus on reproducibility should go hand in hand with the emphasis on software quality. We present a case study in which we identify and fix three bugs in widely used implementations of the state-of-the-art Conformer architecture. Through experiments on speech recognition and translation in various languages, we demonstrate that the presence of bugs does not prevent the achievement of good and reproducible results, which however can lead to incorrect conclusions that potentially misguide future research. As a countermeasure, we propose a Code-quality Checklist and release pangoliNN, a library dedicated to testing neural models, with the goal of promoting coding best practices and improving research software quality within the NLP community.

Renato Sortino, Thomas Cecconello, Andrea DeMarco et al.

Along with the nearing completion of the Square Kilometre Array (SKA), comes an increasing demand for accurate and reliable automated solutions to extract valuable information from the vast amount of data it will allow acquiring. Automated source finding is a particularly important task in this context, as it enables the detection and classification of astronomical objects. Deep-learning-based object detection and semantic segmentation models have proven to be suitable for this purpose. However, training such deep networks requires a high volume of labeled data, which is not trivial to obtain in the context of radio astronomy. Since data needs to be manually labeled by experts, this process is not scalable to large dataset sizes, limiting the possibilities of leveraging deep networks to address several tasks. In this work, we propose RADiff, a generative approach based on conditional diffusion models trained over an annotated radio dataset to generate synthetic images, containing radio sources of different morphologies, to augment existing datasets and reduce the problems caused by class imbalances. We also show that it is possible to generate fully-synthetic image-annotation pairs to automatically augment any annotated dataset. We evaluate the effectiveness of this approach by training a semantic segmentation model on a real dataset augmented in two ways: 1) using synthetic images obtained from real masks, and 2) generating images from synthetic semantic masks. We show an improvement in performance when applying augmentation, gaining up to 18% in performance when using real masks and 4% when augmenting with synthetic masks. Finally, we employ this model to generate large-scale radio maps with the objective of simulating Data Challenges.

Lassi Meronen, Martin Trapp, Andrea Pilzer et al.

Dynamic neural networks are a recent technique that promises a remedy for the increasing size of modern deep learning models by dynamically adapting their computational cost to the difficulty of the inputs. In this way, the model can adjust to a limited computational budget. However, the poor quality of uncertainty estimates in deep learning models makes it difficult to distinguish between hard and easy samples. To address this challenge, we present a computationally efficient approach for post-hoc uncertainty quantification in dynamic neural networks. We show that adequately quantifying and accounting for both aleatoric and epistemic uncertainty through a probabilistic treatment of the last layers improves the predictive performance and aids decision-making when determining the computational budget. In the experiments, we show improvements on CIFAR-100, ImageNet, and Caltech-256 in terms of accuracy, capturing uncertainty, and calibration error.

Andrea Pilzer, Yuxin Hou, Niki Loppi et al.

We introduce visual hints expansion for guiding stereo matching to improve generalization. Our work is motivated by the robustness of Visual Inertial Odometry (VIO) in computer vision and robotics, where a sparse and unevenly distributed set of feature points characterizes a scene. To improve stereo matching, we propose to elevate 2D hints to 3D points. These sparse and unevenly distributed 3D visual hints are expanded using a 3D random geometric graph, which enhances the learning and inference process. We evaluate our proposal on multiple widely adopted benchmarks and show improved performance without access to additional sensors other than the image sequence. To highlight practical applicability and symbiosis with visual odometry, we demonstrate how our methods run on embedded hardware.

Arno Solin, Rui Li, Andrea Pilzer

The fusion of camera sensor and inertial data is a leading method for ego-motion tracking in autonomous and smart devices. State estimation techniques that rely on non-linear filtering are a strong paradigm for solving the associated information fusion task. The de facto inference method in this space is the celebrated extended Kalman filter (EKF), which relies on first-order linearizations of both the dynamical and measurement model. This paper takes a critical look at the practical implications and limitations posed by the EKF, especially under faulty visual feature associations and the presence of strong confounding noise. As an alternative, we revisit the assumed density formulation of Bayesian filtering and employ a moment matching (unscented Kalman filtering) approach to both visual-inertial odometry and visual SLAM. Our results highlight important aspects in robustness both in dynamics propagation and visual measurement updates, and we show state-of-the-art results on EuRoC MAV drone data benchmark.

Tsung-Ming Tai, Sofia Casarin, Andrea Pilzer et al.

Anticipating future actions in videos is challenging, as the observed frames provide only evidence of past activities, requiring the inference of latent intentions to predict upcoming actions. Existing transformer-based approaches, which rely on dot-product attention over pixel representations, often lack the high-level semantics necessary to model video sequences for effective action anticipation. As a result, these methods tend to overfit to explicit visual cues present in the past frames, limiting their ability to capture underlying intentions and degrading generalization to unseen samples. To address this, we propose Action-Guided Attention (AGA), an attention mechanism that explicitly leverages predicted action sequences as queries and keys to guide sequence modeling. Our approach fosters the attention module to emphasize relevant moments from the past based on the upcoming activity and combine this information with the current frame embedding via a dedicated gating function. The design of AGA enables post-training analysis of the knowledge discovered from the training set. Experiments on the widely adopted EPIC-Kitchens-100 benchmark demonstrate that AGA generalizes well from validation to unseen test sets. Post-training analysis can further examine the action dependencies captured by the model and the counterfactual evidence it has internalized, offering transparent and interpretable insights into its anticipative predictions.

Pranav Kasela, Marco Braga, Alessandro Ghiotto et al.

In this paper, we present DIETA, a small, decoder-only Transformer model with 0.5 billion parameters, specifically designed and trained for Italian-English machine translation. We collect and curate a large parallel corpus consisting of approximately 207 million Italian-English sentence pairs across diverse domains, including parliamentary proceedings, legal texts, web-crawled content, subtitles, news, literature and 352 million back-translated data using pretrained models. Additionally, we create and release a new small-scale evaluation set, consisting of 450 sentences, based on 2025 WikiNews articles, enabling assessment of translation quality on contemporary text. Comprehensive evaluations show that DIETA achieves competitive performance on multiple Italian-English benchmarks, consistently ranking in the second quartile of a 32-system leaderboard and outperforming most other sub-3B models on four out of five test suites. The training script, trained models, curated corpus, and newly introduced evaluation set are made publicly available, facilitating further research and development in specialized Italian-English machine translation. https://github.com/pkasela/DIETA-Machine-Translation

Jaime Corsetti, Francesco Giuliari, Davide Boscaini et al.

Functionality understanding in 3D, which aims to identify the functional element in a 3D scene to complete an action (e.g., the correct handle to "Open the second drawer of the cabinet near the bed"), is hindered by the scarcity of real-world data due to the substantial effort needed for its collection and annotation. To address this, we introduce SynthFun3D, the first method for task-based 3D scene synthesis. Given the action description, SynthFun3D generates a 3D indoor environment using a furniture asset database with part-level annotation, ensuring the action can be accomplished. It reasons about the action to automatically identify and retrieve the 3D mask of the correct functional element, enabling the inexpensive and large-scale generation of high-quality annotated data. We validate SynthFun3D through user studies, which demonstrate improved scene-prompt coherence compared to other approaches. Our quantitative results further show that the generated data can either replace real data with minor performance loss or supplement real data for improved performance, thereby providing an inexpensive and scalable solution for data-hungry 3D applications. Project page: github.com/tev-fbk/synthfun3d.

Andrea Pilzer, Dan Xu, Mihai Marian Puscas et al.

While recent deep monocular depth estimation approaches based on supervised regression have achieved remarkable performance, costly ground truth annotations are required during training. To cope with this issue, in this paper we present a novel unsupervised deep learning approach for predicting depth maps and show that the depth estimation task can be effectively tackled within an adversarial learning framework. Specifically, we propose a deep generative network that learns to predict the correspondence field i.e. the disparity map between two image views in a calibrated stereo camera setting. The proposed architecture consists of two generative sub-networks jointly trained with adversarial learning for reconstructing the disparity map and organized in a cycle such as to provide mutual constraints and supervision to each other. Extensive experiments on the publicly available datasets KITTI and Cityscapes demonstrate the effectiveness of the proposed model and competitive results with state of the art methods. The code and trained model are available on https://github.com/andrea-pilzer/unsup-stereo-depthGAN.

Gianni Franchi, Dat Nguyen Trong, Nacim Belkhir et al.

Uncertainty quantification in text-to-image (T2I) generative models is crucial for understanding model behavior and improving output reliability. In this paper, we are the first to quantify and evaluate the uncertainty of T2I models with respect to the prompt. Alongside adapting existing approaches designed to measure uncertainty in the image space, we also introduce Prompt-based UNCertainty Estimation for T2I models (PUNC), a novel method leveraging Large Vision-Language Models (LVLMs) to better address uncertainties arising from the semantics of the prompt and generated images. PUNC utilizes a LVLM to caption a generated image, and then compares the caption with the original prompt in the more semantically meaningful text space. PUNC also enables the disentanglement of both aleatoric and epistemic uncertainties via precision and recall, which image-space approaches are unable to do. Extensive experiments demonstrate that PUNC outperforms state-of-the-art uncertainty estimation techniques across various settings. Uncertainty quantification in text-to-image generation models can be used on various applications including bias detection, copyright protection, and OOD detection. We also introduce a comprehensive dataset of text prompts and generation pairs to foster further research in uncertainty quantification for generative models. Our findings illustrate that PUNC not only achieves competitive performance but also enables novel applications in evaluating and improving the trustworthiness of text-to-image models.

Firas Gabetni, Giuseppe Curci, Andrea Pilzer et al.

Uncertainty quantification (UQ) is essential for deploying deep neural networks in safety-critical settings. Although methods like Deep Ensembles achieve strong UQ performance, their high computational and memory costs hinder scalability to large models. We introduce Hydra Ensembles, an efficient transformer-based ensemble that prunes attention heads to create diverse members and merges them via a new multi-head attention with grouped fully-connected layers. This yields a compact model with inference speed close to a single network, matching or surpassing Deep Ensembles in UQ performance without retraining from scratch. We also provide an in-depth analysis of pruning, showing that naive approaches can harm calibration, whereas Hydra Ensembles preserves robust uncertainty. Experiments on image and text classification tasks, with various architectures, show consistent gains over Deep Ensembles. Remarkably, in zero-shot classification on ImageNet-1k, our approach surpasses state of the art methods, even without requiring additional training.

Simone Facchiano, Stefano Saravalle, Matteo Migliarini et al.

Video generative models democratize the creation of visual content through intuitive instruction following, but they also inherit the biases and harmful concepts embedded within their web-scale training data. This inheritance creates a significant risk, as users can readily generate undesirable and even illegal content. This work introduces the first unlearning technique tailored explicitly for video diffusion models to address this critical issue. Our method requires 5 multi-modal prompt pairs only. Each pair contains a "safe" and an "unsafe" example that differ only by the target concept. Averaging their per-layer latent differences produces a "refusal vector", which, once subtracted from the model parameters, neutralizes the unsafe concept. We introduce a novel low-rank factorization approach on the covariance difference of embeddings that yields robust refusal vectors. This isolates the target concept while minimizing collateral unlearning of other semantics, thus preserving the visual quality of the generated video. Our method preserves the model's generation quality while operating without retraining or access to the original training data. By embedding the refusal direction directly into the model's weights, the suppression mechanism becomes inherently more robust against adversarial bypass attempts compared to surface-level input-output filters. In a thorough qualitative and quantitative evaluation, we show that we can neutralize a variety of harmful contents, including explicit nudity, graphic violence, copyrights, and trademarks. Project page: https://www.pinlab.org/video-unlearning.

Gianni Franchi, Olivier Laurent, Maxence Leguéry et al.

Deep Neural Networks (DNNs) are powerful tools for various computer vision tasks, yet they often struggle with reliable uncertainty quantification - a critical requirement for real-world applications. Bayesian Neural Networks (BNN) are equipped for uncertainty estimation but cannot scale to large DNNs that are highly unstable to train. To address this challenge, we introduce the Adaptable Bayesian Neural Network (ABNN), a simple and scalable strategy to seamlessly transform DNNs into BNNs in a post-hoc manner with minimal computational and training overheads. ABNN preserves the main predictive properties of DNNs while enhancing their uncertainty quantification abilities through simple BNN adaptation layers (attached to normalization layers) and a few fine-tuning steps on pre-trained models. We conduct extensive experiments across multiple datasets for image classification and semantic segmentation tasks, and our results demonstrate that ABNN achieves state-of-the-art performance without the computational budget typically associated with ensemble methods.

Andrea Pilzer, Stéphane Lathuilière, Dan Xu et al.

Recent deep monocular depth estimation approaches based on supervised regression have achieved remarkable performance. However, they require costly ground truth annotations during training. To cope with this issue, in this paper we present a novel unsupervised deep learning approach for predicting depth maps. We introduce a new network architecture, named Progressive Fusion Network (PFN), that is specifically designed for binocular stereo depth estimation. This network is based on a multi-scale refinement strategy that combines the information provided by both stereo views. In addition, we propose to stack twice this network in order to form a cycle. This cycle approach can be interpreted as a form of data-augmentation since, at training time, the network learns both from the training set images (in the forward half-cycle) but also from the synthesized images (in the backward half-cycle). The architecture is jointly trained with adversarial learning. Extensive experiments on the publicly available datasets KITTI, Cityscapes and ApolloScape demonstrate the effectiveness of the proposed model which is competitive with other unsupervised deep learning methods for depth prediction.

Mihai Marian Puscas, Dan Xu, Andrea Pilzer et al.

Inspired by the success of adversarial learning, we propose a new end-to-end unsupervised deep learning framework for monocular depth estimation consisting of two Generative Adversarial Networks (GAN), deeply coupled with a structured Conditional Random Field (CRF) model. The two GANs aim at generating distinct and complementary disparity maps and at improving the generation quality via exploiting the adversarial learning strategy. The deep CRF coupling model is proposed to fuse the generative and discriminative outputs from the dual GAN nets. As such, the model implicitly constructs mutual constraints on the two network branches and between the generator and discriminator. This facilitates the optimization of the whole network for better disparity generation. Extensive experiments on the KITTI, Cityscapes, and Make3D datasets clearly demonstrate the effectiveness of the proposed approach and show superior performance compared to state of the art methods. The code and models are available at https://github.com/mihaipuscas/ 3dv---coupled-crf-disparity.

Zhenyu Zhang, Stéphane Lathuilière, Andrea Pilzer et al.

In this work, we tackle the problem of online adaptation for stereo depth estimation, that consists in continuously adapting a deep network to a target video recordedin an environment different from that of the source training set. To address this problem, we propose a novel Online Meta-Learning model with Adaption (OMLA). Our proposal is based on two main contributions. First, to reducethe domain-shift between source and target feature distributions we introduce an online feature alignment procedurederived from Batch Normalization. Second, we devise a meta-learning approach that exploits feature alignment forfaster convergence in an online learning setting. Additionally, we propose a meta-pre-training algorithm in order toobtain initial network weights on the source dataset whichfacilitate adaptation on future data streams. Experimentally, we show that both OMLA and meta-pre-training helpthe model to adapt faster to a new environment. Our proposal is evaluated on the wellestablished KITTI dataset,where we show that our online method is competitive withstate of the art algorithms trained in a batch setting.

Andrea Pilzer, Stéphane Lathuilière, Nicu Sebe et al.

Nowadays, the majority of state of the art monocular depth estimation techniques are based on supervised deep learning models. However, collecting RGB images with associated depth maps is a very time consuming procedure. Therefore, recent works have proposed deep architectures for addressing the monocular depth prediction task as a reconstruction problem, thus avoiding the need of collecting ground-truth depth. Following these works, we propose a novel self-supervised deep model for estimating depth maps. Our framework exploits two main strategies: refinement via cycle-inconsistency and distillation. Specifically, first a \emph{student} network is trained to predict a disparity map such as to recover from a frame in a camera view the associated image in the opposite view. Then, a backward cycle network is applied to the generated image to re-synthesize back the input image, estimating the opposite disparity. A third network exploits the inconsistency between the original and the reconstructed input frame in order to output a refined depth map. Finally, knowledge distillation is exploited, such as to transfer information from the refinement network to the student. Our extensive experimental evaluation demonstrate the effectiveness of the proposed framework which outperforms state of the art unsupervised methods on the KITTI benchmark.

Xavier Alameda-Pineda, Andrea Pilzer, Dan Xu et al.

In our overly-connected world, the automatic recognition of virality - the quality of an image or video to be rapidly and widely spread in social networks - is of crucial importance, and has recently awaken the interest of the computer vision community. Concurrently, recent progress in deep learning architectures showed that global pooling strategies allow the extraction of activation maps, which highlight the parts of the image most likely to contain instances of a certain class. We extend this concept by introducing a pooling layer that learns the size of the support area to be averaged: the learned top-N average (LENA) pooling. We hypothesize that the latent concepts (feature maps) describing virality may require such a rich pooling strategy. We assess the effectiveness of the LENA layer by appending it on top of a convolutional siamese architecture and evaluate its performance on the task of predicting and localizing virality. We report experiments on two publicly available datasets annotated for virality and show that our method outperforms state-of-the-art approaches.