Gianni Franchi

Xuanlong Yu, Gianni Franchi, Jindong Gu et al. · deepmind, oxford

Uncertainty quantification is critical for deploying deep neural networks (DNNs) in real-world applications. An Auxiliary Uncertainty Estimator (AuxUE) is one of the most effective means to estimate the uncertainty of the main task prediction without modifying the main task model. To be considered robust, an AuxUE must be capable of maintaining its performance and triggering higher uncertainties while encountering Out-of-Distribution (OOD) inputs, i.e., to provide robust aleatoric and epistemic uncertainty. However, for vision regression tasks, current AuxUE designs are mainly adopted for aleatoric uncertainty estimates, and AuxUE robustness has not been explored. In this work, we propose a generalized AuxUE scheme for more robust uncertainty quantification on regression tasks. Concretely, to achieve a more robust aleatoric uncertainty estimation, different distribution assumptions are considered for heteroscedastic noise, and Laplace distribution is finally chosen to approximate the prediction error. For epistemic uncertainty, we propose a novel solution named Discretization-Induced Dirichlet pOsterior (DIDO), which models the Dirichlet posterior on the discretized prediction error. Extensive experiments on age estimation, monocular depth estimation, and super-resolution tasks show that our proposed method can provide robust uncertainty estimates in the face of noisy inputs and that it can be scalable to both image-level and pixel-wise tasks. Code is available at https://github.com/ENSTA-U2IS/DIDO .

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.

Marwane Hariat, Olivier Laurent, Rémi Kazmierczak et al.

Semantic segmentation methods have advanced significantly. Still, their robustness to real-world perturbations and object types not seen during training remains a challenge, particularly in safety-critical applications. We propose a novel approach to improve the robustness of semantic segmentation techniques by leveraging the synergy between label-to-image generators and image-to-label segmentation models. Specifically, we design Robusta, a novel robust conditional generative adversarial network to generate realistic and plausible perturbed images that can be used to train reliable segmentation models. We conduct in-depth studies of the proposed generative model, assess the performance and robustness of the downstream segmentation network, and demonstrate that our approach can significantly enhance the robustness in the face of real-world perturbations, distribution shifts, and out-of-distribution samples. Our results suggest that this approach could be valuable in safety-critical applications, where the reliability of perception modules such as semantic segmentation is of utmost importance and comes with a limited computational budget in inference. We release our code at https://github.com/ENSTA-U2IS-AI/robusta.

Olivier Laurent, Adrien Lafage, Enzo Tartaglione et al.

Deep Ensembles (DE) are a prominent approach for achieving excellent performance on key metrics such as accuracy, calibration, uncertainty estimation, and out-of-distribution detection. However, hardware limitations of real-world systems constrain to smaller ensembles and lower-capacity networks, significantly deteriorating their performance and properties. We introduce Packed-Ensembles (PE), a strategy to design and train lightweight structured ensembles by carefully modulating the dimension of their encoding space. We leverage grouped convolutions to parallelize the ensemble into a single shared backbone and forward pass to improve training and inference speeds. PE is designed to operate within the memory limits of a standard neural network. Our extensive research indicates that PE accurately preserves the properties of DE, such as diversity, and performs equally well in terms of accuracy, calibration, out-of-distribution detection, and robustness to distribution shift. We make our code available at https://github.com/ENSTA-U2IS/torch-uncertainty.

Gianni Franchi, Xuanlong Yu, Andrei Bursuc et al.

Predictive uncertainty estimation is essential for deploying Deep Neural Networks in real-world autonomous systems. However, most successful approaches are computationally intensive. In this work, we attempt to address these challenges in the context of autonomous driving perception tasks. Recently proposed Deterministic Uncertainty Methods (DUM) can only partially meet such requirements as their scalability to complex computer vision tasks is not obvious. In this work we advance a scalable and effective DUM for high-resolution semantic segmentation, that relaxes the Lipschitz constraint typically hindering practicality of such architectures. We learn a discriminant latent space by leveraging a distinction maximization layer over an arbitrarily-sized set of trainable prototypes. Our approach achieves competitive results over Deep Ensembles, the state-of-the-art for uncertainty prediction, on image classification, segmentation and monocular depth estimation tasks. Our code is available at https://github.com/ENSTA-U2IS/LDU

Mouïn Ben Ammar, Nacim Belkhir, Sebastian Popescu et al.

Detecting out-of-distribution (OOD) data is a critical challenge in machine learning due to model overconfidence, often without awareness of their epistemological limits. We hypothesize that ``neural collapse'', a phenomenon affecting in-distribution data for models trained beyond loss convergence, also influences OOD data. To benefit from this interplay, we introduce NECO, a novel post-hoc method for OOD detection, which leverages the geometric properties of ``neural collapse'' and of principal component spaces to identify OOD data. Our extensive experiments demonstrate that NECO achieves state-of-the-art results on both small and large-scale OOD detection tasks while exhibiting strong generalization capabilities across different network architectures. Furthermore, we provide a theoretical explanation for the effectiveness of our method in OOD detection. Code is available at https://gitlab.com/drti/neco

Adrien Bennetot, Gianni Franchi, Javier Del Ser et al.

Although Deep Neural Networks (DNNs) have great generalization and prediction capabilities, their functioning does not allow a detailed explanation of their behavior. Opaque deep learning models are increasingly used to make important predictions in critical environments, and the danger is that they make and use predictions that cannot be justified or legitimized. Several eXplainable Artificial Intelligence (XAI) methods that separate explanations from machine learning models have emerged, but have shortcomings in faithfulness to the model actual functioning and robustness. As a result, there is a widespread agreement on the importance of endowing Deep Learning models with explanatory capabilities so that they can themselves provide an answer to why a particular prediction was made. First, we address the problem of the lack of universal criteria for XAI by formalizing what an explanation is. We also introduced a set of axioms and definitions to clarify XAI from a mathematical perspective. Finally, we present the Greybox XAI, a framework that composes a DNN and a transparent model thanks to the use of a symbolic Knowledge Base (KB). We extract a KB from the dataset and use it to train a transparent model (i.e., a logistic regression). An encoder-decoder architecture is trained on RGB images to produce an output similar to the KB used by the transparent model. Once the two models are trained independently, they are used compositionally to form an explainable predictive model. We show how this new architecture is accurate and explainable in several datasets.

Kai Xu, Rongyu Chen, Gianni Franchi et al.

The capacity of a modern deep learning system to determine if a sample falls within its realm of knowledge is fundamental and important. In this paper, we offer insights and analyses of recent state-of-the-art out-of-distribution (OOD) detection methods - extremely simple activation shaping (ASH). We demonstrate that activation pruning has a detrimental effect on OOD detection, while activation scaling enhances it. Moreover, we propose SCALE, a simple yet effective post-hoc network enhancement method for OOD detection, which attains state-of-the-art OOD detection performance without compromising in-distribution (ID) accuracy. By integrating scaling concepts into the training process to capture a sample's ID characteristics, we propose Intermediate Tensor SHaping (ISH), a lightweight method for training time OOD detection enhancement. We achieve AUROC scores of +1.85\% for near-OOD and +0.74\% for far-OOD datasets on the OpenOOD v1.5 ImageNet-1K benchmark. Our code and models are available at https://github.com/kai422/SCALE.

Gianni Franchi, Marwane Hariat, Xuanlong Yu et al.

Current deep neural networks (DNNs) for autonomous driving computer vision are typically trained on specific datasets that only involve a single type of data and urban scenes. Consequently, these models struggle to handle new objects, noise, nighttime conditions, and diverse scenarios, which is essential for safety-critical applications. Despite ongoing efforts to enhance the resilience of computer vision DNNs, progress has been sluggish, partly due to the absence of benchmarks featuring multiple modalities. We introduce a novel and versatile dataset named InfraParis that supports multiple tasks across three modalities: RGB, depth, and infrared. We assess various state-of-the-art baseline techniques, encompassing models for the tasks of semantic segmentation, object detection, and depth estimation. More visualizations and the download link for InfraParis are available at \href{https://ensta-u2is.github.io/infraParis/}{https://ensta-u2is.github.io/infraParis/}.

Firas Gabetni, Alexandre Rocchi Henry, Nacim Belkhir et al.

Detecting covariate shift is critical for building reliable vision systems. While most prior work focuses on improving robustness to shift, explicitly detecting covariate shift remains underexplored. Existing approaches typically rely on fully supervised training, requiring labeled examples from both original and shifted distributions, which is often impractical. In this paper, we show that covariate shift detection can be effectively addressed with weaker supervision using Positive Unlabeled (PU) learning. However, under covariate shift, in distribution and shifted data overlap significantly, making classical PU methods unstable and sensitive to noise. To overcome this challenge, we introduce Spectral PU Neighborhood Annotation (SPUNA), a geometry aware framework that progressively discovers shifted data by leveraging the local manifold structure of visual features. Extensive experiments show that SPUNA achieves state of the art performance in PU settings and remarkably matches the performances of fully supervised methods. Moreover, our approach transfers robustly across different types of shifts, demonstrating strong generalization capabilities.

Gianni Franchi, Xuanlong Yu, Andrei Bursuc et al.

Predictive uncertainty estimation is essential for safe deployment of Deep Neural Networks in real-world autonomous systems. However, disentangling the different types and sources of uncertainty is non trivial for most datasets, especially since there is no ground truth for uncertainty. In addition, while adverse weather conditions of varying intensities can disrupt neural network predictions, they are usually under-represented in both training and test sets in public datasets.We attempt to mitigate these setbacks and introduce the MUAD dataset (Multiple Uncertainties for Autonomous Driving), consisting of 10,413 realistic synthetic images with diverse adverse weather conditions (night, fog, rain, snow), out-of-distribution objects, and annotations for semantic segmentation, depth estimation, object, and instance detection. MUAD allows to better assess the impact of different sources of uncertainty on model performance. We conduct a thorough experimental study of this impact on several baseline Deep Neural Networks across multiple tasks, and release our dataset to allow researchers to benchmark their algorithm methodically in adverse conditions. More visualizations and the download link for MUAD are available at https://muad-dataset.github.io/.

Denis Mbey Akola, Gianni Franchi

This paper presents a new approach for training two-stage object detection ensemble models, more specifically, Faster R-CNN models to estimate uncertainty. We propose training one Region Proposal Network(RPN) and multiple Fast R-CNN prediction heads is all you need to build a robust deep ensemble network for estimating uncertainty in object detection. We present this approach and provide experiments to show that this approach is much faster than the naive method of fully training all $n$ models in an ensemble. We also estimate the uncertainty by measuring this ensemble model's Expected Calibration Error (ECE). We then further compare the performance of this model with that of Gaussian YOLOv3, a variant of YOLOv3 that models uncertainty using predicted bounding box coordinates. The source code is released at \url{https://github.com/Akola-Mbey-Denis/EfficientEnsemble}

Joseph Hoche, David Brellmann, Gianni Franchi

Uncertainty quantification (UQ) remains a critical challenge in Large Vision Language Models (LVLMs) for reliable predictions and real-world deployment. However, most existing methods are adapted from the LLM literature and primarily focus on the language modality, leaving the contribution of visual information to LVLM uncertainty largely underexplored. In this paper, we investigate how LVLMs process visual information and whether this process can be used to improve uncertainty estimation. By analyzing hidden representations after the integration of visual features during the generation process, we observe that high-confidence predictions rely more heavily on visual content than uncertain ones. Building on this insight, we propose Visual-Grounded Token UQ (VIG-TUQ), a training-free framework that explicitly incorporates visual grounding into uncertainty estimation by weighting token-level language uncertainty with visual grounding scores. We evaluate VIG-TUQ on multiple datasets and across diverse LVLM architectures, including early-fusion, late-fusion, and native-fusion models. Results indicate that our method often improves upon existing token-level uncertainty approaches. Code and data will be made available upon acceptance.

Olivier Laurent, Emanuel Aldea, Gianni Franchi

The distribution of the weights of modern deep neural networks (DNNs) - crucial for uncertainty quantification and robustness - is an eminently complex object due to its extremely high dimensionality. This paper proposes one of the first large-scale explorations of the posterior distribution of deep Bayesian Neural Networks (BNNs), expanding its study to real-world vision tasks and architectures. Specifically, we investigate the optimal approach for approximating the posterior, analyze the connection between posterior quality and uncertainty quantification, delve into the impact of modes on the posterior, and explore methods for visualizing the posterior. Moreover, we uncover weight-space symmetries as a critical aspect for understanding the posterior. To this extent, we develop an in-depth assessment of the impact of both permutation and scaling symmetries that tend to obfuscate the Bayesian posterior. While the first type of transformation is known for duplicating modes, we explore the relationship between the latter and L2 regularization, challenging previous misconceptions. Finally, to help the community improve our understanding of the Bayesian posterior, we will shortly release the first large-scale checkpoint dataset, including thousands of real-world models and our codes.

Adrien Lafage, Olivier Laurent, Firas Gabetni et al.

Deep Neural Networks (DNNs) have demonstrated remarkable performance across various domains, including computer vision and natural language processing. However, they often struggle to accurately quantify the uncertainty of their predictions, limiting their broader adoption in critical real-world applications. Uncertainty Quantification (UQ) for Deep Learning seeks to address this challenge by providing methods to improve the reliability of uncertainty estimates. Although numerous techniques have been proposed, a unified tool offering a seamless workflow to evaluate and integrate these methods remains lacking. To bridge this gap, we introduce Torch-Uncertainty, a PyTorch and Lightning-based framework designed to streamline DNN training and evaluation with UQ techniques and metrics. In this paper, we outline the foundational principles of our library and present comprehensive experimental results that benchmark a diverse set of UQ methods across classification, segmentation, and regression tasks. Our library is available at https://github.com/ENSTA-U2IS-AI/Torch-Uncertainty

Clement Laroudie, Andrei Bursuc, Mai Lan Ha et al.

This paper examines the robustness of a multi-modal computer vision model, CLIP (Contrastive Language-Image Pretraining), in the context of unsupervised learning. The main objective is twofold: first, to evaluate the robustness of CLIP, and second, to explore strategies for augmenting its robustness. To achieve this, we introduce a novel approach named LP-CLIP. This technique involves the distillation of CLIP features through the incorporation of a linear probing layer positioned atop its encoding structure. This newly added layer is trained utilizing pseudo-labels produced by CLIP, coupled with a self-training strategy. The LP-CLIP technique offers a promising approach to enhance the robustness of CLIP without the need for annotations. By leveraging a simple linear probing layer, we aim to improve the model's ability to withstand various uncertainties and challenges commonly encountered in real-world scenarios. Importantly, our approach does not rely on annotated data, which makes it particularly valuable in situations where labeled data might be scarce or costly to obtain. Our proposed approach increases the robustness of CLIP with SOTA results compared to supervised technique on various datasets.

Joseph Hoche, Andrei Bursuc, David Brellmann et al.

Large Vision-Language Models (LVLMs) often produce plausible but unreliable outputs, making robust uncertainty estimation essential. Recent work on semantic uncertainty estimates relies on external models to cluster multiple sampled responses and measure their semantic consistency. However, these clustering methods are often fragile, highly sensitive to minor phrasing variations, and can incorrectly group or separate semantically similar answers, leading to unreliable uncertainty estimates. We propose Semantic Gaussian Process Uncertainty (SGPU), a Bayesian framework that quantifies semantic uncertainty by analyzing the geometric structure of answer embeddings, avoiding brittle clustering. SGPU maps generated answers into a dense semantic space, computes the Gram matrix of their embeddings, and summarizes their semantic configuration via the eigenspectrum. This spectral representation is then fed into a Gaussian Process Classifier that learns to map patterns of semantic consistency to predictive uncertainty, and that can be applied in both black-box and white-box settings. Across six LLMs and LVLMs on eight datasets spanning VQA, image classification, and textual QA, SGPU consistently achieves state-of-the-art calibration (ECE) and discriminative (AUROC, AUARC) performance. We further show that SGPU transfers across models and modalities, indicating that its spectral representation captures general patterns of semantic uncertainty.

Rémi Kazmierczak, Eloïse Berthier, Goran Frehse et al.

In this paper, we introduce an explainable algorithm designed from a multi-modal foundation model, that performs fast and explainable image classification. Drawing inspiration from CLIP-based Concept Bottleneck Models (CBMs), our method creates a latent space where each neuron is linked to a specific word. Observing that this latent space can be modeled with simple distributions, we use a Mixture of Gaussians (MoG) formalism to enhance the interpretability of this latent space. Then, we introduce CLIP-QDA, a classifier that only uses statistical values to infer labels from the concepts. In addition, this formalism allows for both local and global explanations. These explanations come from the inner design of our architecture, our work is part of a new family of greybox models, combining performances of opaque foundation models and the interpretability of transparent models. Our empirical findings show that in instances where the MoG assumption holds, CLIP-QDA achieves similar accuracy with state-of-the-art methods CBMs. Our explanations compete with existing XAI methods while being faster to compute.

Alexandre Rocchi--Henry, Thomas Fel, Gianni Franchi

Two traditions of interpretability have evolved side by side but seldom spoken to each other: Concept Bottleneck Models (CBMs), which prescribe what a concept should be, and Sparse Autoencoders (SAEs), which discover what concepts emerge. While CBMs use supervision to align activations with human-labeled concepts, SAEs rely on sparse coding to uncover emergent ones. We show that both paradigms instantiate the same geometric structure: each learns a set of linear directions in activation space whose nonnegative combinations form a concept cone. Supervised and unsupervised methods thus differ not in kind but in how they select this cone. Building on this view, we propose an operational bridge between the two paradigms. CBMs provide human-defined reference geometries, while SAEs can be evaluated by how well their learned cones approximate or contain those of CBMs. This containment framework yields quantitative metrics linking inductive biases -- such as SAE type, sparsity, or expansion ratio -- to emergence of plausible\footnote{We adopt the terminology of \citet{jacovi2020towards}, who distinguish between faithful explanations (accurately reflecting model computations) and plausible explanations (aligning with human intuition and domain knowledge). CBM concepts are plausible by construction -- selected or annotated by humans -- though not necessarily faithful to the true latent factors that organise the data manifold.} concepts. Using these metrics, we uncover a ``sweet spot'' in both sparsity and expansion factor that maximizes both geometric and semantic alignment with CBM concepts. Overall, our work unifies supervised and unsupervised concept discovery through a shared geometric framework, providing principled metrics to measure SAE progress and assess how well discovered concept align with plausible human concepts.

Galadrielle Humblot-Renaux, Gianni Franchi, Sergio Escalera et al.

Out-of-distribution (OOD) detection is an important building block in trustworthy image recognition systems as unknown classes may arise at test-time. OOD detection methods typically revolve around a single classifier, leading to a split in the research field between the classical supervised setting (e.g. ResNet18 classifier trained on CIFAR100) vs. the zero-shot setting (class names fed as prompts to CLIP). In both cases, an overarching challenge is that the OOD detection performance is implicitly constrained by the classifier's capabilities on in-distribution (ID) data. In this work, we show that given a little open-mindedness from both ends, remarkable OOD detection can be achieved by instead creating a heterogeneous ensemble - COOkeD combines the predictions of a closed-world classifier trained end-to-end on a specific dataset, a zero-shot CLIP classifier, and a linear probe classifier trained on CLIP image features. While bulky at first sight, this approach is modular, post-hoc and leverages the availability of pre-trained VLMs, thus introduces little overhead compared to training a single standard classifier. We evaluate COOkeD on popular CIFAR100 and ImageNet benchmarks, but also consider more challenging, realistic settings ranging from training-time label noise, to test-time covariate shift, to zero-shot shift which has been previously overlooked. Despite its simplicity, COOkeD achieves state-of-the-art performance and greater robustness compared to both classical and CLIP-based OOD detection methods. Code is available at https://github.com/glhr/COOkeD

Rémi Kazmierczak, Gianni Franchi, Nacim Belkhir et al.

Several metrics exist to quantify the similarity between images, but they are inefficient when it comes to measure the similarity of highly distorted images. In this work, we propose to empirically investigate perceptual metrics based on deep neural networks for tackling the Image Quality Assessment (IQA) task. We study deep perceptual metrics according to different hyperparameters like the network's architecture or training procedure. Finally, we propose our multi-resolution perceptual metric (MR-Perceptual), that allows us to aggregate perceptual information at different resolutions and outperforms standard perceptual metrics on IQA tasks with varying image deformations. Our code is available at https://github.com/ENSTA-U2IS/MR_perceptual

Rémi Kazmierczak, Eloïse Berthier, Goran Frehse et al.

As artificial intelligence systems become increasingly integrated into daily life, the field of explainability has gained significant attention. This trend is particularly driven by the complexity of modern AI models and their decision-making processes. The advent of foundation models, characterized by their extensive generalization capabilities and emergent uses, has further complicated this landscape. Foundation models occupy an ambiguous position in the explainability domain: their complexity makes them inherently challenging to interpret, yet they are increasingly leveraged as tools to construct explainable models. In this survey, we explore the intersection of foundation models and eXplainable AI (XAI) in the vision domain. We begin by compiling a comprehensive corpus of papers that bridge these fields. Next, we categorize these works based on their architectural characteristics. We then discuss the challenges faced by current research in integrating XAI within foundation models. Furthermore, we review common evaluation methodologies for these combined approaches. Finally, we present key observations and insights from our survey, offering directions for future research in this rapidly evolving field.

Adrien Lafage, Mathieu Barbier, Gianni Franchi et al.

Accurate trajectory forecasting is crucial for the performance of various systems, such as advanced driver-assistance systems and self-driving vehicles. These forecasts allow us to anticipate events that lead to collisions and, therefore, to mitigate them. Deep Neural Networks have excelled in motion forecasting, but overconfidence and weak uncertainty quantification persist. Deep Ensembles address these concerns, yet applying them to multimodal distributions remains challenging. In this paper, we propose a novel approach named Hierarchical Light Transformer Ensembles (HLT-Ens) aimed at efficiently training an ensemble of Transformer architectures using a novel hierarchical loss function. HLT-Ens leverages grouped fully connected layers, inspired by grouped convolution techniques, to capture multimodal distributions effectively. We demonstrate that HLT-Ens achieves state-of-the-art performance levels through extensive experimentation, offering a promising avenue for improving trajectory forecasting techniques.

Marwane Hariat, Gianni Franchi, David Filliat et al.

We present SS3D, a web-scale SfM-based self-supervision pretraining pipeline for feed-forward 3D estimation from monocular video. Our model jointly predicts depth, ego-motion, and intrinsics in a single forward pass and is trained/evaluated as a coherent end-to-end 3D estimator. To stabilize joint learning, we use an intrinsics-first two-stage schedule and a unified single-checkpoint evaluation protocol. Scaling SfM self-supervision to unconstrained web video is challenging due to weak multi-view observability and strong corpus heterogeneity; we address these with a multi-view signal proxy (MVS) used for filtering and curriculum sampling, and with expert training distilled into a single student. Pretraining on YouTube-8M (~100M frames after filtering) yields strong cross-domain zero-shot transfer and improved fine-tuning performance over prior self-supervised baselines. We release the pretrained checkpoint and code.

Rémi Kazmierczak, Steve Azzolin, Eloïse Berthier et al.

We introduce PASTA (Perceptual Assessment System for explanaTion of Artificial Intelligence), a novel human-centric framework for evaluating eXplainable AI (XAI) techniques in computer vision. Our first contribution is the creation of the PASTA-dataset, the first large-scale benchmark that spans a diverse set of models and both saliency-based and concept-based explanation methods. This dataset enables robust, comparative analysis of XAI techniques based on human judgment. Our second contribution is an automated, data-driven benchmark that predicts human preferences using the PASTA-dataset. This scoring called PASTA-score method offers scalable, reliable, and consistent evaluation aligned with human perception. Additionally, our benchmark allows for comparisons between explanations across different modalities, an aspect previously unaddressed. We then propose to apply our scoring method to probe the interpretability of existing models and to build more human interpretable XAI methods.

Mingxuan Liu, Zhun Zhong, Jun Li et al.

In this work, we introduce and study the novel task of Open-ended Semantic Multiple Clustering (OpenSMC). Given a large, unstructured image collection, the goal is to automatically discover several, diverse semantic clustering criteria (e.g., Activity or Location) from the images, and subsequently organize them according to the discovered criteria, without requiring any human input. Our framework, X-Cluster: eXploratory Clustering, treats text as a reasoning proxy: it concurrently scans the entire image collection, proposes candidate criteria in natural language, and groups images into meaningful clusters per criterion. This radically differs from previous works, which either assume predefined clustering criteria or fixed cluster counts. To evaluate X-Cluster, we create two new benchmarks, COCO-4C and Food-4C, each annotated with four distinct grouping criteria and corresponding cluster labels. Experiments show that X-Cluster can effectively reveal meaningful partitions on several datasets. Finally, we use X-Cluster to achieve various real-world applications, including uncovering hidden biases in text-to-image (T2I) generative models and analyzing image virality on social media. Project page: https://oatmealliu.github.io/xcluster.html

Mouïn Ben Ammar, Arturo Mendoza, Nacim Belkhir et al.

In line with the development of deep learning, this survey examines the transformative role of Transformers and foundation models in advancing visual anomaly detection (VAD). We explore how these architectures, with their global receptive fields and adaptability, address challenges such as long-range dependency modeling, contextual modeling and data scarcity. The survey categorizes VAD methods into reconstruction-based, feature-based and zero/few-shot approaches, highlighting the paradigm shift brought about by foundation models. By integrating attention mechanisms and leveraging large-scale pre-training, Transformers and foundation models enable more robust, interpretable, and scalable anomaly detection solutions. This work provides a comprehensive review of state-of-the-art techniques, their strengths, limitations, and emerging trends in leveraging these architectures for VAD.

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.

Francesco Taioli, Edoardo Zorzi, Gianni Franchi et al.

Language-driven instance object navigation assumes that human users initiate the task by providing a detailed description of the target instance to the embodied agent. While this description is crucial for distinguishing the target from visually similar instances in a scene, providing it prior to navigation can be demanding for human. To bridge this gap, we introduce Collaborative Instance object Navigation (CoIN), a new task setting where the agent actively resolve uncertainties about the target instance during navigation in natural, template-free, open-ended dialogues with human. We propose a novel training-free method, Agent-user Interaction with UncerTainty Awareness (AIUTA), which operates independently from the navigation policy, and focuses on the human-agent interaction reasoning with Vision-Language Models (VLMs) and Large Language Models (LLMs). First, upon object detection, a Self-Questioner model initiates a self-dialogue within the agent to obtain a complete and accurate observation description with a novel uncertainty estimation technique. Then, an Interaction Trigger module determines whether to ask a question to the human, continue or halt navigation, minimizing user input. For evaluation, we introduce CoIN-Bench, with a curated dataset designed for challenging multi-instance scenarios. CoIN-Bench supports both online evaluation with humans and reproducible experiments with simulated user-agent interactions. On CoIN-Bench, we show that AIUTA serves as a competitive baseline, while existing language-driven instance navigation methods struggle in complex multi-instance scenes. Code and benchmark will be available upon acceptance at https://intelligolabs.github.io/CoIN/

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.

Rémi Kazmierczak, Steve Azzolin, Eloïse Berthier et al.

This paper addresses explainable AI (XAI) through the lens of Concept Bottleneck Models (CBMs) that do not require explicit concept annotations, relying instead on concepts extracted using CLIP in a zero-shot manner. We show that CLIP, which is central in these techniques, is prone to concept hallucination, incorrectly predicting the presence or absence of concepts within an image in scenarios used in numerous CBMs, hence undermining the faithfulness of explanations. To mitigate this issue, we introduce Concept Hallucination Inhibition via Localized Interpretability (CHILI), a technique that disentangles image embeddings and localizes pixels corresponding to target concepts. Furthermore, our approach supports the generation of saliency-based explanations that are more interpretable.

Gaetan Brison, Soobash Daiboo, Samy Aimeur et al.

We introduce FakeParts, a new class of deepfakes characterized by subtle, localized manipulations to specific spatial regions or temporal segments of otherwise authentic videos. Unlike fully synthetic content, these partial manipulations, ranging from altered facial expressions to object substitutions and background modifications, blend seamlessly with real elements, making them particularly deceptive and difficult to detect. To address the critical gap in detection capabilities, we present FakePartsBench, the first large-scale benchmark dataset specifically designed to capture the full spectrum of partial deepfakes. Comprising over 25K videos with pixel-level and frame-level manipulation annotations, our dataset enables comprehensive evaluation of detection methods. Our user studies demonstrate that FakeParts reduces human detection accuracy by over 30% compared to traditional deepfakes, with similar performance degradation observed in state-of-the-art detection models. This work identifies an urgent vulnerability in current deepfake detection approaches and provides the necessary resources to develop more robust methods for partial video manipulations.

Sofiia Chorna, Kateryna Tarelkina, Eloïse Berthier et al.

While concept-based interpretability methods have traditionally focused on local explanations of neural network predictions, we propose a novel framework and interactive tool that extends these methods into the domain of mechanistic interpretability. Our approach enables a global dissection of model behavior by analyzing how high-level semantic attributes (referred to as concepts) emerge, interact, and propagate through internal model components. Unlike prior work that isolates individual neurons or predictions, our framework systematically quantifies how semantic concepts are represented across layers, revealing latent circuits and information flow that underlie model decision-making. A key innovation is our visualization platform that we named BAGEL (for Bias Analysis with a Graph for global Explanation Layers), which presents these insights in a structured knowledge graph, allowing users to explore concept-class relationships, identify spurious correlations, and enhance model trustworthiness. Our framework is model-agnostic, scalable, and contributes to a deeper understanding of how deep learning models generalize (or fail to) in the presence of dataset biases. The demonstration is available at https://knowledge-graph-ui-4a7cb5.gitlab.io/.

Mouïn Ben Ammar, David Brellmann, Arturo Mendoza et al.

Out-of-distribution (OOD) detection is essential for ensuring the reliability and safety of machine learning systems. In recent years, it has received increasing attention, particularly through post-hoc detection and training-based methods. In this paper, we focus on post-hoc OOD detection, which enables identifying OOD samples without altering the model's training procedure or objective. Our primary goal is to investigate the relationship between model capacity and its OOD detection performance. Specifically, we aim to answer the following question: Does the Double Descent phenomenon manifest in post-hoc OOD detection? This question is crucial, as it can reveal whether overparameterization, which is already known to benefit generalization, can also enhance OOD detection. Despite the growing interest in these topics by the classic supervised machine learning community, this intersection remains unexplored for OOD detection. We empirically demonstrate that the Double Descent effect does indeed appear in post-hoc OOD detection. Furthermore, we provide theoretical insights to explain why this phenomenon emerges in such setting. Finally, we show that the overparameterized regime does not yield superior results consistently, and we propose a method to identify the optimal regime for OOD detection based on our observations.

Guoxuan Xia, Olivier Laurent, Gianni Franchi et al.

Label smoothing (LS) is a popular regularisation method for training neural networks as it is effective in improving test accuracy and is simple to implement. ``Hard'' one-hot labels are ``smoothed'' by uniformly distributing probability mass to other classes, reducing overfitting. Prior work has suggested that in some cases LS can degrade selective classification (SC) -- where the aim is to reject misclassifications using a model's uncertainty. In this work, we first demonstrate empirically across an extended range of large-scale tasks and architectures that LS consistently degrades SC. We then address a gap in existing knowledge, providing an explanation for this behaviour by analysing logit-level gradients: LS degrades the uncertainty rank ordering of correct vs incorrect predictions by suppressing the max logit more when a prediction is likely to be correct, and less when it is likely to be wrong. This elucidates previously reported experimental results where strong classifiers underperform in SC. We then demonstrate the empirical effectiveness of post-hoc logit normalisation for recovering lost SC performance caused by LS. Furthermore, linking back to our gradient analysis, we again provide an explanation for why such normalisation is effective.

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.

Xuanlong Yu, Gianni Franchi, Emanuel Aldea

Monocular depth is important in many tasks, such as 3D reconstruction and autonomous driving. Deep learning based models achieve state-of-the-art performance in this field. A set of novel approaches for estimating monocular depth consists of transforming the regression task into a classification one. However, there is a lack of detailed descriptions and comparisons for Classification Approaches for Regression (CAR) in the community and no in-depth exploration of their potential for uncertainty estimation. To this end, this paper will introduce a taxonomy and summary of CAR approaches, a new uncertainty estimation solution for CAR, and a set of experiments on depth accuracy and uncertainty quantification for CAR-based models on KITTI dataset. The experiments reflect the differences in the portability of various CAR methods on two backbones. Meanwhile, the newly proposed method for uncertainty estimation can outperform the ensembling method with only one forward propagation.

Xuanlong Yu, Gianni Franchi, Emanuel Aldea

It has become critical for deep learning algorithms to quantify their output uncertainties to satisfy reliability constraints and provide accurate results. Uncertainty estimation for regression has received less attention than classification due to the more straightforward standardized output of the latter class of tasks and their high importance. However, regression problems are encountered in a wide range of applications in computer vision. We propose SLURP, a generic approach for regression uncertainty estimation via a side learner that exploits the output and the intermediate representations generated by the main task model. We test SLURP on two critical regression tasks in computer vision: monocular depth and optical flow estimation. In addition, we conduct exhaustive benchmarks comprising transfer to different datasets and the addition of aleatoric noise. The results show that our proposal is generic and readily applicable to various regression problems and has a low computational cost with respect to existing solutions.

Gianni Franchi, Nacim Belkhir, Mai Lan Ha et al.

Along with predictive performance and runtime speed, reliability is a key requirement for real-world semantic segmentation. Reliability encompasses robustness, predictive uncertainty and reduced bias. To improve reliability, we introduce Superpixel-mix, a new superpixel-based data augmentation method with teacher-student consistency training. Unlike other mixing-based augmentation techniques, mixing superpixels between images is aware of object boundaries, while yielding consistent gains in segmentation accuracy. Our proposed technique achieves state-of-the-art results in semi-supervised semantic segmentation on the Cityscapes dataset. Moreover, Superpixel-mix improves the reliability of semantic segmentation by reducing network uncertainty and bias, as confirmed by competitive results under strong distributions shift (adverse weather, image corruptions) and when facing out-of-distribution data.

Mai Lan Ha, Gianni Franchi, Emanuel Aldea et al.

Discriminative features play an important role in image and object classification and also in other fields of research such as semi-supervised learning, fine-grained classification, out of distribution detection. Inspired by Linear Discriminant Analysis (LDA), we propose an optimization called Neural Discriminant Analysis (NDA) for Deep Convolutional Neural Networks (DCNNs). NDA transforms deep features to become more discriminative and, therefore, improves the performances in various tasks. Our proposed optimization has two primary goals for inter- and intra-class variances. The first one is to minimize variances within each individual class. The second goal is to maximize pairwise distances between features coming from different classes. We evaluate our NDA optimization in different research fields: general supervised classification, fine-grained classification, semi-supervised learning, and out of distribution detection. We achieve performance improvements in all the fields compared to baseline methods that do not use NDA. Besides, using NDA, we also surpass the state of the art on the four tasks on various testing datasets.

Yufei Hu, Nacim Belkhir, Jesus Angulo et al.

Deep Neural Networks (DNNs) are generated by sequentially performing linear and non-linear processes. Using a combination of linear and non-linear procedures is critical for generating a sufficiently deep feature space. The majority of non-linear operators are derivations of activation functions or pooling functions. Mathematical morphology is a branch of mathematics that provides non-linear operators for a variety of image processing problems. We investigate the utility of integrating these operations in an end-to-end deep learning framework in this paper. DNNs are designed to acquire a realistic representation for a particular job. Morphological operators give topological descriptors that convey salient information about the shapes of objects depicted in images. We propose a method based on meta-learning to incorporate morphological operators into DNNs. The learned architecture demonstrates how our novel morphological operations significantly increase DNN performance on various tasks, including picture classification and edge detection.

Natalia Díaz-Rodríguez, Alberto Lamas, Jules Sanchez et al.

The latest Deep Learning (DL) models for detection and classification have achieved an unprecedented performance over classical machine learning algorithms. However, DL models are black-box methods hard to debug, interpret, and certify. DL alone cannot provide explanations that can be validated by a non technical audience. In contrast, symbolic AI systems that convert concepts into rules or symbols -- such as knowledge graphs -- are easier to explain. However, they present lower generalisation and scaling capabilities. A very important challenge is to fuse DL representations with expert knowledge. One way to address this challenge, as well as the performance-explainability trade-off is by leveraging the best of both streams without obviating domain expert knowledge. We tackle such problem by considering the symbolic knowledge is expressed in form of a domain expert knowledge graph. We present the eXplainable Neural-symbolic learning (X-NeSyL) methodology, designed to learn both symbolic and deep representations, together with an explainability metric to assess the level of alignment of machine and human expert explanations. The ultimate objective is to fuse DL representations with expert domain knowledge during the learning process to serve as a sound basis for explainability. X-NeSyL methodology involves the concrete use of two notions of explanation at inference and training time respectively: 1) EXPLANet: Expert-aligned eXplainable Part-based cLAssifier NETwork Architecture, a compositional CNN that makes use of symbolic representations, and 2) SHAP-Backprop, an explainable AI-informed training procedure that guides the DL process to align with such symbolic representations in form of knowledge graphs. We showcase X-NeSyL methodology using MonuMAI dataset for monument facade image classification, and demonstrate that our approach improves explainability and performance.

Gianni Franchi, Andrei Bursuc, Emanuel Aldea et al.

Bayesian neural networks (BNNs) have been long considered an ideal, yet unscalable solution for improving the robustness and the predictive uncertainty of deep neural networks. While they could capture more accurately the posterior distribution of the network parameters, most BNN approaches are either limited to small networks or rely on constraining assumptions such as parameter independence. These drawbacks have enabled prominence of simple, but computationally heavy approaches such as Deep Ensembles, whose training and testing costs increase linearly with the number of networks. In this work we aim for efficient deep BNNs amenable to complex computer vision architectures, e.g. ResNet50 DeepLabV3+, and tasks, e.g. semantic segmentation, with fewer assumptions on the parameters. We achieve this by leveraging variational autoencoders (VAEs) to learn the interaction and the latent distribution of the parameters at each network layer. Our approach, Latent-Posterior BNN (LP-BNN), is compatible with the recent BatchEnsemble method, leading to highly efficient ({in terms of computation and} memory during both training and testing) ensembles. LP-BNN s attain competitive results across multiple metrics in several challenging benchmarks for image classification, semantic segmentation and out-of-distribution detection.

Gianni Franchi, Andrei Bursuc, Emanuel Aldea et al.

Deep neural networks (DNNs) are powerful learning models yet their results are not always reliable. This is due to the fact that modern DNNs are usually uncalibrated and we cannot characterize their epistemic uncertainty. In this work, we propose a new technique to quantify the epistemic uncertainty of data easily. This method consists in mixing the predictions of an ensemble of DNNs trained to classify One class vs All the other classes (OVA) with predictions from a standard DNN trained to perform All vs All (AVA) classification. On the one hand, the adjustment provided by the AVA DNN to the score of the base classifiers allows for a more fine-grained inter-class separation. On the other hand, the two types of classifiers enforce mutually their detection of out-of-distribution (OOD) samples, circumventing entirely the requirement of using such samples during training. Our method achieves state of the art performance in quantifying OOD data across multiple datasets and architectures while requiring little hyper-parameter tuning.

Gianni Franchi, Andrei Bursuc, Emanuel Aldea et al.

During training, the weights of a Deep Neural Network (DNN) are optimized from a random initialization towards a nearly optimum value minimizing a loss function. Only this final state of the weights is typically kept for testing, while the wealth of information on the geometry of the weight space, accumulated over the descent towards the minimum is discarded. In this work we propose to make use of this knowledge and leverage it for computing the distributions of the weights of the DNN. This can be further used for estimating the epistemic uncertainty of the DNN by sampling an ensemble of networks from these distributions. To this end we introduce a method for tracking the trajectory of the weights during optimization, that does not require any changes in the architecture nor on the training procedure. We evaluate our method on standard classification and regression benchmarks, and on out-of-distribution detection for classification and semantic segmentation. We achieve competitive results, while preserving computational efficiency in comparison to other popular approaches.

Gianni Franchi, Angela Yao, Andreas Kolb

We propose a novel single-image super-resolution approach based on the geostatistical method of kriging. Kriging is a zero-bias minimum-variance estimator that performs spatial interpolation based on a weighted average of known observations. Rather than solving for the kriging weights via the traditional method of inverting covariance matrices, we propose a supervised form in which we learn a deep network to generate said weights. We combine the kriging weight generation and kriging process into a joint network that can be learned end-to-end. Our network achieves competitive super-resolution results as other state-of-the-art methods. In addition, since the super-resolution process follows a known statistical framework, we are able to estimate bias and variance, something which is rarely possible for other deep networks.

Gianni Franchi, Jesus Angulo, Dino Sejdinovic

We propose a novel approach for pixel classification in hyperspectral images, leveraging on both the spatial and spectral information in the data. The introduced method relies on a recently proposed framework for learning on distributions -- by representing them with mean elements in reproducing kernel Hilbert spaces (RKHS) and formulating a classification algorithm therein. In particular, we associate each pixel to an empirical distribution of its neighbouring pixels, a judicious representation of which in an RKHS, in conjunction with the spectral information contained in the pixel itself, give a new explicit set of features that can be fed into a suite of standard classification techniques -- we opt for a well-established framework of support vector machines (SVM). Furthermore, the computational complexity is reduced via random Fourier features formalism. We study the consistency and the convergence rates of the proposed method and the experiments demonstrate strong performance on hyperspectral data with gains in comparison to the state-of-the-art results.