Soufiane Belharbi

Dimitrios Kollias, Panagiotis Tzirakis, Alan Cowen et al.

The 10th Affective & Behavior Analysis in-the-Wild (ABAW) Workshop and Competition, held at CVPR 2026, continues to advance research on modelling, analysis, understanding of human affect and behavior in real-world, unconstrained environments. The workshop maintains its dual structure, comprising both a competition and a paper track. The ABAW Competition introduces a diverse set of challenges targeting key aspects of affective and behavioral understanding, including continuous affect (valence-arousal) estimation, discrete affect (expression and action unit) recognition, as well as more complex behavior analysis tasks, such as emotional mimicry intensity estimation, ambivalence/hesitancy recognition and fine-grained violence detection. These challenges are built upon large-scale in-the-wild datasets, providing comprehensive benchmarks for state-of-the-art approaches. In parallel, the paper track presents a wide range of contributions spanning pose, motion & behavior estimation, affect modelling & multimodal learning, benchmarks, datasets & evaluation protocols, fairness, robustness & deployment. Overall, the 10th ABAW Workshop and Competition continues to serve as a key platform for benchmarking, collaboration and innovation, shaping the development of next-generation multimodal, human-centered AI systems.

Nicolas Richet, Soufiane Belharbi, Haseeb Aslam et al.

Systems for multimodal emotion recognition (ER) are commonly trained to extract features from different modalities (e.g., visual, audio, and textual) that are combined to predict individual basic emotions. However, compound emotions often occur in real-world scenarios, and the uncertainty of recognizing such complex emotions over diverse modalities is challenging for feature-based models. As an alternative, emerging large language models (LLMs) like BERT and LLaMA can rely on explicit non-verbal cues that may be translated from different non-textual modalities (e.g., audio and visual) into text. Textualization of modalities augments data with emotional cues to help the LLM encode the interconnections between all modalities in a shared text space. In such text-based models, prior knowledge of ER tasks is leveraged to textualize relevant non-verbal cues such as audio tone from vocal expressions, and action unit intensity from facial expressions. Since the pre-trained weights are publicly available for many LLMs, training on large-scale datasets is unnecessary, allowing to fine-tune for downstream tasks such as compound ER (CER). This paper compares the potential of text- and feature-based approaches for compound multimodal ER in videos. Experiments were conducted on the challenging C-EXPR-DB dataset in the wild for CER, and contrasted with results on the MELD dataset for basic ER. Our results indicate that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild. However, higher accuracy can be achieved when the video data has rich transcripts. Our code is available.

Soufiane Belharbi, Ismail Ben Ayed, Luke McCaffrey et al.

Weakly supervised video object localization (WSVOL) allows locating object in videos using only global video tags such as object class. State-of-art methods rely on multiple independent stages, where initial spatio-temporal proposals are generated using visual and motion cues, then prominent objects are identified and refined. Localization is done by solving an optimization problem over one or more videos, and video tags are typically used for video clustering. This requires a model per-video or per-class making for costly inference. Moreover, localized regions are not necessary discriminant because of unsupervised motion methods like optical flow, or because video tags are discarded from optimization. In this paper, we leverage the successful class activation mapping (CAM) methods, designed for WSOL based on still images. A new Temporal CAM (TCAM) method is introduced to train a discriminant deep learning (DL) model to exploit spatio-temporal information in videos, using an aggregation mechanism, called CAM-Temporal Max Pooling (CAM-TMP), over consecutive CAMs. In particular, activations of regions of interest (ROIs) are collected from CAMs produced by a pretrained CNN classifier to build pixel-wise pseudo-labels for training the DL model. In addition, a global unsupervised size constraint, and local constraint such as CRF are used to yield more accurate CAMs. Inference over single independent frames allows parallel processing of a clip of frames, and real-time localization. Extensive experiments on two challenging YouTube-Objects datasets for unconstrained videos, indicate that CAM methods (trained on independent frames) can yield decent localization accuracy. Our proposed TCAM method achieves a new state-of-art in WSVOL accuracy, and visual results suggest that it can be adapted for subsequent tasks like visual object tracking and detection. Code is publicly available.

Soufiane Belharbi, Shakeeb Murtaza, Marco Pedersoli et al.

Leveraging spatiotemporal information in videos is critical for weakly supervised video object localization (WSVOL) tasks. However, state-of-the-art methods only rely on visual and motion cues, while discarding discriminative information, making them susceptible to inaccurate localizations. Recently, discriminative models have been explored for WSVOL tasks using a temporal class activation mapping (CAM) method. Although their results are promising, objects are assumed to have limited movement from frame to frame, leading to degradation in performance for relatively long-term dependencies. This paper proposes a novel CAM method for WSVOL that exploits spatiotemporal information in activation maps during training without constraining an object's position. Its training relies on Co-Localization, hence, the name CoLo-CAM. Given a sequence of frames, localization is jointly learned based on color cues extracted across the corresponding maps, by assuming that an object has similar color in consecutive frames. CAM activations are constrained to respond similarly over pixels with similar colors, achieving co-localization. This improves localization performance because the joint learning creates direct communication among pixels across all image locations and over all frames, allowing for transfer, aggregation, and correction of localizations. Co-localization is integrated into training by minimizing the color term of a conditional random field (CRF) loss over a sequence of frames/CAMs. Extensive experiments on two challenging YouTube-Objects datasets of unconstrained videos show the merits of our method, and its robustness to long-term dependencies, leading to new state-of-the-art performance for WSVOL task.

Shakeeb Murtaza, Soufiane Belharbi, Marco Pedersoli et al.

Drones are employed in a growing number of visual recognition applications. A recent development in cell tower inspection is drone-based asset surveillance, where the autonomous flight of a drone is guided by localizing objects of interest in successive aerial images. In this paper, we propose a method to train deep weakly-supervised object localization (WSOL) models based only on image-class labels to locate object with high confidence. To train our localizer, pseudo labels are efficiently harvested from a self-supervised vision transformers (SSTs). However, since SSTs decompose the scene into multiple maps containing various object parts, and do not rely on any explicit supervisory signal, they cannot distinguish between the object of interest and other objects, as required WSOL. To address this issue, we propose leveraging the multiple maps generated by the different transformer heads to acquire pseudo-labels for training a deep WSOL model. In particular, a new Discriminative Proposals Sampling (DiPS) method is introduced that relies on a CNN classifier to identify discriminative regions. Then, foreground and background pixels are sampled from these regions in order to train a WSOL model for generating activation maps that can accurately localize objects belonging to a specific class. Empirical results on the challenging TelDrone dataset indicate that our proposed approach can outperform state-of-art methods over a wide range of threshold values over produced maps. We also computed results on CUB dataset, showing that our method can be adapted for other tasks.

Shakeeb Murtaza, Soufiane Belharbi, Marco Pedersoli et al.

Self-supervised vision transformers can generate accurate localization maps of the objects in an image. However, since they decompose the scene into multiple maps containing various objects, and they do not rely on any explicit supervisory signal, they cannot distinguish between the object of interest from other objects, as required in weakly-supervised object localization (WSOL). To address this issue, we propose leveraging the multiple maps generated by the different transformer heads to acquire pseudo-labels for training a WSOL model. In particular, a new discriminative proposals sampling method is introduced that relies on a pretrained CNN classifier to identify discriminative regions. Then, foreground and background pixels are sampled from these regions in order to train a WSOL model for generating activation maps that can accurately localize objects belonging to a specific class. Empirical results on the challenging CUB benchmark dataset indicate that our proposed approach can outperform state-of-art methods over a wide range of threshold values. Our method provides class activation maps with a better coverage of foreground object regions w.r.t. the background.

Soufiane Belharbi, Jérôme Rony, Jose Dolz et al.

Trained using only image class label, deep weakly supervised methods allow image classification and ROI segmentation for interpretability. Despite their success on natural images, they face several challenges over histology data where ROI are visually similar to background making models vulnerable to high pixel-wise false positives. These methods lack mechanisms for modeling explicitly non-discriminative regions which raises false-positive rates. We propose novel regularization terms, which enable the model to seek both non-discriminative and discriminative regions, while discouraging unbalanced segmentations and using only image class label. Our method is composed of two networks: a localizer that yields segmentation mask, followed by a classifier. The training loss pushes the localizer to build a segmentation mask that holds most discrimiantive regions while simultaneously modeling background regions. Comprehensive experiments over two histology datasets showed the merits of our method in reducing false positives and accurately segmenting ROI.

Shakeeb Murtaza, Soufiane Belharbi, Marco Pedersoli et al.

Self-supervised vision transformers (SSTs) have shown great potential to yield rich localization maps that highlight different objects in an image. However, these maps remain class-agnostic since the model is unsupervised. They often tend to decompose the image into multiple maps containing different objects while being unable to distinguish the object of interest from background noise objects. In this paper, Discriminative Pseudo-label Sampling (DiPS) is introduced to leverage these class-agnostic maps for weakly-supervised object localization (WSOL), where only image-class labels are available. Given multiple attention maps, DiPS relies on a pre-trained classifier to identify the most discriminative regions of each attention map. This ensures that the selected ROIs cover the correct image object while discarding the background ones, and, as such, provides a rich pool of diverse and discriminative proposals to cover different parts of the object. Subsequently, these proposals are used as pseudo-labels to train our new transformer-based WSOL model designed to perform classification and localization tasks. Unlike standard WSOL methods, DiPS optimizes performance in both tasks by using a transformer encoder and a dedicated output head for each task, each trained using dedicated loss functions. To avoid overfitting a single proposal and promote better object coverage, a single proposal is randomly selected among the top ones for a training image at each training step. Experimental results on the challenging CUB, ILSVRC, OpenImages, and TelDrone datasets indicate that our architecture, in combination with our transformer-based proposals, can yield better localization performance than state-of-the-art methods.

Manuela González-González, Soufiane Belharbi, Muhammad Osama Zeeshan et al.

Using behavioural science, health interventions focus on behaviour change by providing a framework to help patients acquire and maintain healthy habits that improve medical outcomes. In-person interventions are costly and difficult to scale, especially in resource-limited regions. Digital health interventions offer a cost-effective approach, potentially supporting independent living and self-management. Automating such interventions, especially through machine learning, has gained considerable attention recently. Ambivalence and hesitancy (A/H) play a primary role for individuals to delay, avoid, or abandon health interventions. A/H are subtle and conflicting emotions that place a person in a state between positive and negative evaluations of a behaviour, or between acceptance and refusal to engage in it. They manifest as affective inconsistency across modalities or within a modality, such as language, facial, vocal expressions, and body language. While experts can be trained to recognize A/H, integrating them into digital health interventions is costly and less effective. Automatic A/H recognition is therefore critical for the personalization and cost-effectiveness of digital health interventions. Here, we explore the application of deep learning models for A/H recognition in videos, a multi-modal task by nature. In particular, this paper covers three learning setups: supervised learning, unsupervised domain adaptation for personalization, and zero-shot inference via large language models (LLMs). Our experiments are conducted on the unique and recently published BAH video dataset for A/H recognition. Our results show limited performance, suggesting that more adapted multi-modal models are required for accurate A/H recognition. Better methods for modeling spatio-temporal and multimodal fusion are necessary to leverage conflicts within/across modalities.

Masoumeh Sharafi, Muhammad Osama Zeeshan, Soufiane Belharbi et al.

Facial expression recognition (FER) in videos requires model personalization to capture the considerable variations across subjects. Vision-language models (VLMs) offer strong transfer to downstream tasks through image-text alignment, but their performance can still degrade under inter-subject distribution shifts. Personalizing models using test-time adaptation (TTA) methods can mitigate this challenge. However, most state-of-the-art TTA methods rely on unsupervised parameter optimization, introducing computational overhead that is impractical in many real-world applications. This paper introduces TTA through Cache Personalization (TTA-CaP), a cache-based TTA method that enables cost-effective (gradient-free) personalization of VLMs for video FER. Prior cache-based TTA methods rely solely on dynamic memories that store test samples, which can accumulate errors and drift due to noisy pseudo-labels. TTA-CaP leverages three coordinated caches: a personalized source cache that stores source-domain prototypes, a positive target cache that accumulates reliable subject-specific samples, and a negative target cache that stores low-confidence cases as negative samples to reduce the impact of noisy pseudo-labels. Cache updates and replacement are controlled by a tri-gate mechanism based on temporal stability, confidence, and consistency with the personalized cache. Finally, TTA-CaP refines predictions through fusion of embeddings, yielding refined representations that support temporally stable video-level predictions. Our experiments on three challenging video FER datasets, BioVid, StressID, and BAH, indicate that TTA-CaP can outperform state-of-the-art TTA methods under subject-specific and environmental shifts, while maintaining low computational and memory overhead for real-world deployment.

Banafsheh Karimian, Alexis Guichemerre, Soufiane Belharbi et al.

Breast cancer remains a leading cause of cancer-related mortality worldwide. Longitudinal mammography risk prediction models improve multi-year breast cancer risk prediction based on prior screening exams. However, in real-world clinical practice, longitudinal histories are often incomplete, irregular, or unavailable due to missed screenings, first-time examinations, heterogeneous acquisition schedules, or archival constraints. The absence of prior exams degrades the performance of longitudinal risk models and limits their practical applicability. While substantial longitudinal history is available during training, prior exams are commonly absent at test time. In this paper, we address missing history at inference time and propose a longitudinal risk prediction method that uses mammography history as privileged information during training and distills its prognostic value into a student model that only requires the current exam at inference time. The key idea is a privileged multi-teacher distillation scheme with horizon-specific teachers: each teacher is trained on the full longitudinal history to specialize in one prediction horizon, while the student receives only a reconstructed history derived from the current exam. This allows the student to inherit horizon-dependent longitudinal risk cues without requiring prior screening exams at deployment. Our new Privileged History Distillation (PHD) method is validated on a large longitudinal mammography dataset with multi-year cancer outcomes, CSAW-CC, comparing full-history and no-history baselines to their distilled counterparts. Using time-dependent AUC across horizons, our privileged history distillation method markedly improves the performance of long-horizon prediction over no-history models and is comparable to that of full-history models, while using only the current exam at inference time.

Soufiane Belharbi, Mara KM Whitford, Phuong Hoang et al.

Confocal fluorescence microscopy is one of the most accessible and widely used imaging techniques for the study of biological processes at the cellular and subcellular levels. Scanning confocal microscopy allows the capture of high-quality images from thick three-dimensional (3D) samples, yet suffers from well-known limitations such as photobleaching and phototoxicity of specimens caused by intense light exposure, limiting its applications. Cellular damage can be alleviated by changing imaging parameters to reduce light exposure, often at the expense of image quality. Machine/deep learning methods for single-image super-resolution (SISR) can be applied to restore image quality by upscaling lower-resolution (LR) images to yield high-resolution images (HR). These SISR methods have been successfully applied to photo-realistic images due partly to the abundance of publicly available data. In contrast, the lack of publicly available data partly limits their application and success in scanning confocal microscopy. In this paper, we introduce a large scanning confocal microscopy dataset named SR-CACO-2 that is comprised of low- and high-resolution image pairs marked for three different fluorescent markers. It allows the evaluation of performance of SISR methods on three different upscaling levels (X2, X4, X8). SR-CACO-2 contains the human epithelial cell line Caco-2 (ATCC HTB-37), and it is composed of 2,200 unique images, captured with four resolutions and three markers, forming 9,937 image patches for SISR methods. We provide benchmarking results for 16 state-of-the-art methods of the main SISR families. Results show that these methods have limited success in producing high-resolution textures. The dataset is freely accessible under a Creative Commons license (CC BY-NC-SA 4.0). Our dataset, code and pretrained weights for SISR methods are available: https://github.com/sbelharbi/sr-caco-2.

Soufiane Belharbi, Ismail Ben Ayed, Luke McCaffrey et al.

We propose a new constrained-optimization formulation for deep ordinal classification, in which uni-modality of the label distribution is enforced implicitly via a set of inequality constraints over all the pairs of adjacent labels. Based on (c-1) constraints for c labels, our model is non-parametric and, therefore, more flexible than the existing deep ordinal classification techniques. Unlike these, it does not restrict the learned representation to a single and specific parametric model (or penalty) imposed on all the labels. Therefore, it enables the training to explore larger spaces of solutions, while removing the need for ad hoc choices and scaling up to large numbers of labels. It can be used in conjunction with any standard classification loss and any deep architecture. To tackle the ensuing challenging optimization problem, we solve a sequence of unconstrained losses based on a powerful extension of the log-barrier method. This handles effectively competing constraints and accommodates standard SGD for deep networks, while avoiding computationally expensive Lagrangian dual steps and outperforming substantially penalty methods. Furthermore, we propose a new performance metric for ordinal classification, as a proxy to measure distribution uni-modality, referred to as the Sides Order Index (SOI). We report comprehensive evaluations and comparisons to state-of-the-art methods on benchmark public datasets for several ordinal classification tasks, showing the merits of our approach in terms of label consistency, classification accuracy and scalability. Importantly, enforcing label consistency with our model does not incur higher classification errors, unlike many existing ordinal classification methods. A public reproducible PyTorch implementation is provided. (https://github.com/sbelharbi/unimodal-prob-deep-oc-free-distribution)

Soufiane Belharbi, Jérôme Rony, Jose Dolz et al.

Pointwise localization allows more precise localization and accurate interpretability, compared to bounding box, in applications where objects are highly unstructured such as in medical domain. In this work, we focus on weakly supervised localization (WSL) where a model is trained to classify an image and localize regions of interest at pixel-level using only global image annotation. Typical convolutional attentions maps are prune to high false positive regions. To alleviate this issue, we propose a new deep learning method for WSL, composed of a localizer and a classifier, where the localizer is constrained to determine relevant and irrelevant regions using conditional entropy (CE) with the aim to reduce false positive regions. Experimental results on a public medical dataset and two natural datasets, using Dice index, show that, compared to state of the art WSL methods, our proposal can provide significant improvements in terms of image-level classification and pixel-level localization (low false positive) with robustness to overfitting. A public reproducible PyTorch implementation is provided in: https://github.com/sbelharbi/wsol-min-max-entropy-interpretability .

Soufiane Belharbi, Clément Chatelain, Romain Hérault et al.

Training deep neural networks is known to require a large number of training samples. However, in many applications only few training samples are available. In this work, we tackle the issue of training neural networks for classification task when few training samples are available. We attempt to solve this issue by proposing a new regularization term that constrains the hidden layers of a network to learn class-wise invariant representations. In our regularization framework, learning invariant representations is generalized to the class membership where samples with the same class should have the same representation. Numerical experiments over MNIST and its variants showed that our proposal helps improving the generalization of neural network particularly when trained with few samples. We provide the source code of our framework https://github.com/sbelharbi/learning-class-invariant-features .

Soufiane Belharbi, Romain Hérault, Clément Chatelain et al.

A deep neural network model is a powerful framework for learning representations. Usually, it is used to learn the relation $x \to y$ by exploiting the regularities in the input $x$. In structured output prediction problems, $y$ is multi-dimensional and structural relations often exist between the dimensions. The motivation of this work is to learn the output dependencies that may lie in the output data in order to improve the prediction accuracy. Unfortunately, feedforward networks are unable to exploit the relations between the outputs. In order to overcome this issue, we propose in this paper a regularization scheme for training neural networks for these particular tasks using a multi-task framework. Our scheme aims at incorporating the learning of the output representation $y$ in the training process in an unsupervised fashion while learning the supervised mapping function $x \to y$. We evaluate our framework on a facial landmark detection problem which is a typical structured output task. We show over two public challenging datasets (LFPW and HELEN) that our regularization scheme improves the generalization of deep neural networks and accelerates their training. The use of unlabeled data and label-only data is also explored, showing an additional improvement of the results. We provide an opensource implementation (https://github.com/sbelharbi/structured-output-ae) of our framework.

Paul Waligora, Haseeb Aslam, Osama Zeeshan et al.

Multimodal emotion recognition (MMER) systems typically outperform unimodal systems by leveraging the inter- and intra-modal relationships between, e.g., visual, textual, physiological, and auditory modalities. This paper proposes an MMER method that relies on a joint multimodal transformer (JMT) for fusion with key-based cross-attention. This framework can exploit the complementary nature of diverse modalities to improve predictive accuracy. Separate backbones capture intra-modal spatiotemporal dependencies within each modality over video sequences. Subsequently, our JMT fusion architecture integrates the individual modality embeddings, allowing the model to effectively capture inter- and intra-modal relationships. Extensive experiments on two challenging expression recognition tasks -- (1) dimensional emotion recognition on the Affwild2 dataset (with face and voice) and (2) pain estimation on the Biovid dataset (with face and biosensors) -- indicate that our JMT fusion can provide a cost-effective solution for MMER. Empirical results show that MMER systems with our proposed fusion allow us to outperform relevant baseline and state-of-the-art methods.

Soufiane Belharbi, Marco Pedersoli, Alessandro Lameiras Koerich et al.

Although state-of-the-art classifiers for facial expression recognition (FER) can achieve a high level of accuracy, they lack interpretability, an important feature for end-users. Experts typically associate spatial action units (\aus) from a codebook to facial regions for the visual interpretation of expressions. In this paper, the same expert steps are followed. A new learning strategy is proposed to explicitly incorporate \au cues into classifier training, allowing to train deep interpretable models. During training, this \au codebook is used, along with the input image expression label, and facial landmarks, to construct a \au heatmap that indicates the most discriminative image regions of interest w.r.t the facial expression. This valuable spatial cue is leveraged to train a deep interpretable classifier for FER. This is achieved by constraining the spatial layer features of a classifier to be correlated with \au heatmaps. Using a composite loss, the classifier is trained to correctly classify an image while yielding interpretable visual layer-wise attention correlated with \au maps, simulating the expert decision process. Our strategy only relies on image class expression for supervision, without additional manual annotations. Our new strategy is generic, and can be applied to any deep CNN- or transformer-based classifier without requiring any architectural change or significant additional training time. Our extensive evaluation on two public benchmarks \rafdb, and \affectnet datasets shows that our proposed strategy can improve layer-wise interpretability without degrading classification performance. In addition, we explore a common type of interpretable classifiers that rely on class activation mapping (CAM) methods, and show that our approach can also improve CAM interpretability.

Muhammad Haseeb Aslam, Muhammad Osama Zeeshan, Soufiane Belharbi et al.

Deep learning models for multimodal expression recognition have reached remarkable performance in controlled laboratory environments because of their ability to learn complementary and redundant semantic information. However, these models struggle in the wild, mainly because of the unavailability and quality of modalities used for training. In practice, only a subset of the training-time modalities may be available at test time. Learning with privileged information enables models to exploit data from additional modalities that are only available during training. State-of-the-art knowledge distillation (KD) methods have been proposed to distill information from multiple teacher models (each trained on a modality) to a common student model. These privileged KD methods typically utilize point-to-point matching, yet have no explicit mechanism to capture the structural information in the teacher representation space formed by introducing the privileged modality. Experiments were performed on two challenging problems - pain estimation on the Biovid dataset (ordinal classification) and arousal-valance prediction on the Affwild2 dataset (regression). Results show that our proposed method can outperform state-of-the-art privileged KD methods on these problems. The diversity among modalities and fusion architectures indicates that PKDOT is modality- and model-agnostic.

Alexis Guichemerre, Soufiane Belharbi, Tsiry Mayet et al.

Given the emergence of deep learning, digital pathology has gained popularity for cancer diagnosis based on histology images. Deep weakly supervised object localization (WSOL) models can be trained to classify histology images according to cancer grade and identify regions of interest (ROIs) for interpretation, using inexpensive global image-class annotations. A WSOL model initially trained on some labeled source image data can be adapted using unlabeled target data in cases of significant domain shifts caused by variations in staining, scanners, and cancer type. In this paper, we focus on source-free (unsupervised) domain adaptation (SFDA), a challenging problem where a pre-trained source model is adapted to a new target domain without using any source domain data for privacy and efficiency reasons. SFDA of WSOL models raises several challenges in histology, most notably because they are not intended to adapt for both classification and localization tasks. In this paper, 4 state-of-the-art SFDA methods, each one representative of a main SFDA family, are compared for WSOL in terms of classification and localization accuracy. They are the SFDA-Distribution Estimation, Source HypOthesis Transfer, Cross-Domain Contrastive Learning, and Adaptively Domain Statistics Alignment. Experimental results on the challenging Glas (smaller, breast cancer) and Camelyon16 (larger, colon cancer) histology datasets indicate that these SFDA methods typically perform poorly for localization after adaptation when optimized for classification.

Muhammad Osama Zeeshan, Muhammad Haseeb Aslam, Soufiane Belharbi et al.

Adapting a deep learning model to a specific target individual is a challenging facial expression recognition (FER) task that may be achieved using unsupervised domain adaptation (UDA) methods. Although several UDA methods have been proposed to adapt deep FER models across source and target data sets, multiple subject-specific source domains are needed to accurately represent the intra- and inter-person variability in subject-based adaption. This paper considers the setting where domains correspond to individuals, not entire datasets. Unlike UDA, multi-source domain adaptation (MSDA) methods can leverage multiple source datasets to improve the accuracy and robustness of the target model. However, previous methods for MSDA adapt image classification models across datasets and do not scale well to a more significant number of source domains. This paper introduces a new MSDA method for subject-based domain adaptation in FER. It efficiently leverages information from multiple source subjects (labeled source domain data) to adapt a deep FER model to a single target individual (unlabeled target domain data). During adaptation, our subject-based MSDA first computes a between-source discrepancy loss to mitigate the domain shift among data from several source subjects. Then, a new strategy is employed to generate augmented confident pseudo-labels for the target subject, allowing a reduction in the domain shift between source and target subjects. Experiments performed on the challenging BioVid heat and pain dataset with 87 subjects and the UNBC-McMaster shoulder pain dataset with 25 subjects show that our subject-based MSDA can outperform state-of-the-art methods yet scale well to multiple subject-based source domains.

Masoumeh Sharafi, Emma Ollivier, Muhammad Osama Zeeshan et al.

Facial Expression Recognition (FER) from videos is a crucial task in various application areas, such as human-computer interaction and health diagnosis and monitoring (e.g., assessing pain and depression). Beyond the challenges of recognizing subtle emotional or health states, the effectiveness of deep FER models is often hindered by the considerable inter-subject variability in expressions. Source-free (unsupervised) domain adaptation (SFDA) methods may be employed to adapt a pre-trained source model using only unlabeled target domain data, thereby avoiding data privacy, storage, and transmission issues. Typically, SFDA methods adapt to a target domain dataset corresponding to an entire population and assume it includes data from all recognition classes. However, collecting such comprehensive target data can be difficult or even impossible for FER in healthcare applications. In many real-world scenarios, it may be feasible to collect a short neutral control video (which displays only neutral expressions) from target subjects before deployment. These videos can be used to adapt a model to better handle the variability of expressions among subjects. This paper introduces the Disentangled SFDA (DSFDA) method to address the challenge posed by adapting models with missing target expression data. DSFDA leverages data from a neutral target control video for end-to-end generation and adaptation of target data with missing non-neutral data. Our method learns to disentangle features related to expressions and identity while generating the missing non-neutral expression data for the target subject, thereby enhancing model accuracy. Additionally, our self-supervision strategy improves model adaptation by reconstructing target images that maintain the same identity and source expression.

Shakeeb Murtaza, Soufiane Belharbi, Marco Pedersoli et al.

Weakly Supervised Object Localization (WSOL) allows training deep learning models for classification and localization (LOC) using only global class-level labels. The absence of bounding box (bbox) supervision during training raises challenges in the literature for hyper-parameter tuning, model selection, and evaluation. WSOL methods rely on a validation set with bbox annotations for model selection, and a test set with bbox annotations for threshold estimation for producing bboxes from localization maps. This approach, however, is not aligned with the WSOL setting as these annotations are typically unavailable in real-world scenarios. Our initial empirical analysis shows a significant decline in LOC performance when model selection and threshold estimation rely solely on class labels and the image itself, respectively, compared to using manual bbox annotations. This highlights the importance of incorporating bbox labels for optimal model performance. In this paper, a new WSOL evaluation protocol is proposed that provides LOC information without the need for manual bbox annotations. In particular, we generated noisy pseudo-boxes from a pretrained off-the-shelf region proposal method such as Selective Search, CLIP, and RPN for model selection. These bboxes are also employed to estimate the threshold from LOC maps, circumventing the need for test-set bbox annotations. Our experiments with several WSOL methods on ILSVRC and CUB datasets show that using the proposed pseudo-bboxes for validation facilitates the model selection and threshold estimation, with LOC performance comparable to those selected using GT bboxes on the validation set and threshold estimation on the test set. It also outperforms models selected using class-level labels, and then dynamically thresholded based solely on LOC maps.

Manuela González-González, Soufiane Belharbi, Muhammad Osama Zeeshan et al.

Recognizing complex emotions linked to ambivalence and hesitancy (A/H) can play a critical role in the personalization and effectiveness of digital behaviour change interventions. These subtle and conflicting emotions are manifested by a discord between multiple modalities, such as facial and vocal expressions, and body language. Although experts can be trained to identify A/H, integrating them into digital interventions is costly and less effective. Automatic learning systems provide a cost-effective alternative that can adapt to individual users, and operate seamlessly within real-time, and resource-limited environments. However, there are currently no datasets available for the design of ML models to recognize A/H. This paper introduces a first Behavioural Ambivalence/Hesitancy (BAH) dataset collected for subject-based multimodal recognition of A/H in videos. It contains videos from 224 participants captured across 9 provinces in Canada, with different age, and ethnicity. Through our web platform, we recruited participants to answer 7 questions, some of which were designed to elicit A/H while recording themselves via webcam with microphone. BAH amounts to 1,118 videos for a total duration of 8.26 hours with 1.5 hours of A/H. Our behavioural team annotated timestamp segments to indicate where A/H occurs, and provide frame- and video-level annotations with the A/H cues. Video transcripts and their timestamps are also included, along with cropped and aligned faces in each frame, and a variety of participants meta-data. We include results baselines for BAH at frame- and video-level recognition in multi-modal setups, in addition to zero-shot prediction, and for personalization using unsupervised domain adaptation. The limited performance of baseline models highlights the challenges of recognizing A/H in real-world videos. The data, code, and pretrained weights are available.

Shakeeb Murtaza, Soufiane Belharbi, Marco Pedersoli et al.

Weakly supervised object localization (WSOL) using classification models trained with only image-class labels remains an important challenge in computer vision. Given their reliance on classification objectives, traditional WSOL methods like class activation mapping focus on the most discriminative object parts, often missing the full spatial extent. In contrast, recent WSOL methods based on vision-language models like CLIP require ground truth classes or external classifiers to produce a localization map, limiting their deployment in downstream tasks. Moreover, methods like GenPromp attempt to address these issues but introduce considerable complexity due to their reliance on conditional denoising processes and intricate prompt learning. This paper introduces Text Distillation for Localization (TeD-Loc), an approach that directly distills knowledge from CLIP text embeddings into the model backbone and produces patch-level localization. Multiple instance learning of these image patches allows for accurate localization and classification using one model without requiring external classifiers. Such integration of textual and visual modalities addresses the longstanding challenge of achieving accurate localization and classification concurrently, as WSOL methods in the literature typically converge at different epochs. Extensive experiments show that leveraging text embeddings and localization cues provides a cost-effective WSOL model. TeD-Loc improves Top-1 LOC accuracy over state-of-the-art models by about 5% on both CUB and ILSVRC datasets, while significantly reducing computational complexity compared to GenPromp.

Alexis Guichemerre, Soufiane Belharbi, Mohammadhadi Shateri et al.

Weakly supervised object localization (WSOL) methods allow training models to classify images and localize ROIs. WSOL only requires low-cost image-class annotations yet provides a visually interpretable classifier. Standard WSOL methods rely on class activation mapping (CAM) methods to produce spatial localization maps according to a single- or two-step strategy. While both strategies have made significant progress, they still face several limitations with histology images. Single-step methods can easily result in under- or over-activation due to the limited visual ROI saliency in histology images and scarce localization cues. They also face the well-known issue of asynchronous convergence between classification and localization tasks. The two-step approach is sub-optimal because it is constrained to a frozen classifier, limiting the capacity for localization. Moreover, these methods also struggle when applied to out-of-distribution (OOD) datasets. In this paper, a multi-task approach for WSOL is introduced for simultaneous training of both tasks to address the asynchronous convergence problem. In particular, localization is performed in the pixel-feature space of an image encoder that is shared with classification. This allows learning discriminant features and accurate delineation of foreground/background regions to support ROI localization and image classification. We propose PixelCAM, a cost-effective foreground/background pixel-wise classifier in the pixel-feature space that allows for spatial object localization. Using partial-cross entropy, PixelCAM is trained using pixel pseudo-labels collected from a pretrained WSOL model. Both image and pixel-wise classifiers are trained simultaneously using standard gradient descent. In addition, our pixel classifier can easily be integrated into CNN- and transformer-based architectures without any modifications.

Alexis Guichemerre, Banafsheh Karimian, Soufiane Belharbi et al.

Weakly Supervised Object Localization (WSOL) models enable joint classification and region-of-interest localization in histology images using only image-class supervision. When deployed in a target domain, distributions shift remains a major cause of performance degradation, especially when applied on new organs or institutions with different staining protocols and scanner characteristics. Under stronger cross-domain shifts, WSOL predictions can become biased toward dominant classes, producing highly skewed pseudo-label distributions in the target domain. Source-Free (Unsupervised) Domain Adaptation (SFDA) methods are commonly employed to address domain shift. However, because they rely on self-training, the initial bias is reinforced over training iterations, degrading both classification and localization tasks. We identify this amplification of prediction bias as a primary obstacle to the SFDA of WSOL models in histopathology. This paper introduces \sfdadep, a method inspired by machine unlearning that formulates SFDA as an iterative process of identifying and correcting prediction bias. It periodically identifies target images from over-predicted classes and selectively reduces the predictive confidence for uncertain (high entropy) images, while preserving confident predictions. This process reduces the drift of decision boundaries and bias toward dominant classes. A jointly optimized pixel-level classifier further restores discriminative localization features under distribution shift. Extensive experiments on cross-organ and -center histopathology benchmarks (glas, CAMELYON-16, CAMELYON-17) with several WSOL models show that SFDA-DeP consistently improves classification and localization over state-of-the-art SFDA baselines. {\small Code: \href{https://anonymous.4open.science/r/SFDA-DeP-1797/}{anonymous.4open.science/r/SFDA-DeP-1797/}}

Masoumeh Sharafi, Soufiane Belharbi, Houssem Ben Salem et al.

Facial expression recognition (FER) models are employed in many video-based affective computing applications, such as human-computer interaction and healthcare monitoring. However, deep FER models often struggle with subtle expressions and high inter-subject variability, limiting their performance in real-world applications. To improve their performance, source-free domain adaptation (SFDA) methods have been proposed to personalize a pretrained source model using only unlabeled target domain data, thereby avoiding data privacy, storage, and transmission constraints. This paper addresses a challenging scenario where source data is unavailable for adaptation, and only unlabeled target data consisting solely of neutral expressions is available. SFDA methods are not typically designed to adapt using target data from only a single class. Further, using models to generate facial images with non-neutral expressions can be unstable and computationally intensive. In this paper, personalized feature translation (PFT) is proposed for SFDA. Unlike current image translation methods for SFDA, our lightweight method operates in the latent space. We first pre-train the translator on the source domain data to transform the subject-specific style features from one source subject into another. Expression information is preserved by optimizing a combination of expression consistency and style-aware objectives. Then, the translator is adapted on neutral target data, without using source data or image synthesis. By translating in the latent space, PFT avoids the complexity and noise of face expression generation, producing discriminative embeddings optimized for classification. Using PFT eliminates the need for image synthesis, reduces computational overhead (using a lightweight translator), and only adapts part of the model, making the method efficient compared to image-based translation.

Soufiane Belharbi, Marco Pedersoli, Ismail Ben Ayed et al.

Using only global image-class labels, weakly-supervised learning methods, such as class activation mapping, allow training CNNs to jointly classify an image, and locate regions of interest associated with the predicted class. However, without any guidance at the pixel level, such methods may yield inaccurate regions. This problem is known to be more challenging with histology images than with natural ones, since objects are less salient, structures have more variations, and foreground and background regions have stronger similarities. Therefore, computer vision methods for visual interpretation of CNNs may not directly apply. In this paper, a simple yet efficient method based on a composite loss is proposed to learn information from the fully negative samples (i.e., samples without positive regions), and thereby reduce false positives/negatives. Our new loss function contains two complementary terms: the first exploits positive evidence collected from the CNN classifier, while the second leverages the fully negative samples from training data. In particular, a pre-trained CNN is equipped with a decoder that allows refining the regions of interest. The CNN is exploited to collect both positive and negative evidence at the pixel level to train the decoder. Our method called NEGEV benefits from the fully negative samples that naturally occur in the data, without any additional supervision signals beyond image-class labels. Extensive experiments show that our proposed method can substantial outperform related state-of-art methods on GlaS (public benchmark for colon cancer), and Camelyon16 (patch-based benchmark for breast cancer using three different backbones). Our results highlight the benefits of using both positive and negative evidence, the first obtained from a classifier, and the other naturally available in datasets.

Soufiane Belharbi, Aydin Sarraf, Marco Pedersoli et al.

Class Activation Mapping (CAM) methods have recently gained much attention for weakly-supervised object localization (WSOL) tasks. They allow for CNN visualization and interpretation without training on fully annotated image datasets. CAM methods are typically integrated within off-the-shelf CNN backbones, such as ResNet50. Due to convolution and pooling operations, these backbones yield low resolution CAMs with a down-scaling factor of up to 32, contributing to inaccurate localizations. Interpolation is required to restore full size CAMs, yet it does not consider the statistical properties of objects, such as color and texture, leading to activations with inconsistent boundaries, and inaccurate localizations. As an alternative, we introduce a generic method for parametric upscaling of CAMs that allows constructing accurate full resolution CAMs (F-CAMs). In particular, we propose a trainable decoding architecture that can be connected to any CNN classifier to produce highly accurate CAM localizations. Given an original low resolution CAM, foreground and background pixels are randomly sampled to fine-tune the decoder. Additional priors such as image statistics and size constraints are also considered to expand and refine object boundaries. Extensive experiments, over three CNN backbones and six WSOL baselines on the CUB-200-2011 and OpenImages datasets, indicate that our F-CAM method yields a significant improvement in CAM localization accuracy. F-CAM performance is competitive with state-of-art WSOL methods, yet it requires fewer computations during inference.

Madhu Kiran, R Gnana Praveen, Le Thanh Nguyen-Meidine et al.

In real-world video surveillance applications, person re-identification (ReID) suffers from the effects of occlusions and detection errors. Despite recent advances, occlusions continue to corrupt the features extracted by state-of-art CNN backbones, and thereby deteriorate the accuracy of ReID systems. To address this issue, methods in the literature use an additional costly process such as pose estimation, where pose maps provide supervision to exclude occluded regions. In contrast, we introduce a novel Holistic Guidance (HG) method that relies only on person identity labels, and on the distribution of pairwise matching distances of datasets to alleviate the problem of occlusion, without requiring additional supervision. Hence, our proposed student-teacher framework is trained to address the occlusion problem by matching the distributions of between- and within-class distances (DCDs) of occluded samples with that of holistic (non-occluded) samples, thereby using the latter as a soft labeled reference to learn well separated DCDs. This approach is supported by our empirical study where the distribution of between- and within-class distances between images have more overlap in occluded than holistic datasets. In particular, features extracted from both datasets are jointly learned using the student model to produce an attention map that allows separating visible regions from occluded ones. In addition to this, a joint generative-discriminative backbone is trained with a denoising autoencoder, allowing the system to self-recover from occlusions. Extensive experiments on several challenging public datasets indicate that the proposed approach can outperform state-of-the-art methods on both occluded and holistic datasets

Soufiane Belharbi, Jérôme Rony, Jose Dolz et al.

Weakly-supervised learning (WSL) has recently triggered substantial interest as it mitigates the lack of pixel-wise annotations. Given global image labels, WSL methods yield pixel-level predictions (segmentations), which enable to interpret class predictions. Despite their recent success, mostly with natural images, such methods can face important challenges when the foreground and background regions have similar visual cues, yielding high false-positive rates in segmentations, as is the case in challenging histology images. WSL training is commonly driven by standard classification losses, which implicitly maximize model confidence, and locate the discriminative regions linked to classification decisions. Therefore, they lack mechanisms for modeling explicitly non-discriminative regions and reducing false-positive rates. We propose novel regularization terms, which enable the model to seek both non-discriminative and discriminative regions, while discouraging unbalanced segmentations. We introduce high uncertainty as a criterion to localize non-discriminative regions that do not affect classifier decision, and describe it with original Kullback-Leibler (KL) divergence losses evaluating the deviation of posterior predictions from the uniform distribution. Our KL terms encourage high uncertainty of the model when the latter inputs the latent non-discriminative regions. Our loss integrates: (i) a cross-entropy seeking a foreground, where model confidence about class prediction is high; (ii) a KL regularizer seeking a background, where model uncertainty is high; and (iii) log-barrier terms discouraging unbalanced segmentations. Comprehensive experiments and ablation studies over the public GlaS colon cancer data and a Camelyon16 patch-based benchmark for breast cancer show substantial improvements over state-of-the-art WSL methods, and confirm the effect of our new regularizers.

Soufiane Belharbi, Ismail Ben Ayed, Luke McCaffrey et al.

CNN visualization and interpretation methods, like class-activation maps (CAMs), are typically used to highlight the image regions linked to class predictions. These models allow to simultaneously classify images and extract class-dependent saliency maps, without the need for costly pixel-level annotations. However, they typically yield segmentations with high false-positive rates and, therefore, coarse visualisations, more so when processing challenging images, as encountered in histology. To mitigate this issue, we propose an active learning (AL) framework, which progressively integrates pixel-level annotations during training. Given training data with global image-level labels, our deep weakly-supervised learning model jointly performs supervised image-level classification and active learning for segmentation, integrating pixel annotations by an oracle. Unlike standard AL methods that focus on sample selection, we also leverage large numbers of unlabeled images via pseudo-segmentations (i.e., self-learning at the pixel level), and integrate them with the oracle-annotated samples during training. We report extensive experiments over two challenging benchmarks -- high-resolution medical images (histology GlaS data for colon cancer) and natural images (CUB-200-2011 for bird species). Our results indicate that, by simply using random sample selection, the proposed approach can significantly outperform state-of the-art CAMs and AL methods, with an identical oracle-supervision budget. Our code is publicly available.

Akhil Meethal, Marco Pedersoli, Soufiane Belharbi et al.

Weakly supervised object localization is a challenging task in which the object of interest should be localized while learning its appearance. State-of-the-art methods recycle the architecture of a standard CNN by using the activation maps of the last layer for localizing the object. While this approach is simple and works relatively well, object localization relies on different features than classification, thus, a specialized localization mechanism is required during training to improve performance. In this paper, we propose a convolutional, multi-scale spatial localization network that provides accurate localization for the object of interest. Experimental results on CUB-200-2011 and ImageNet datasets show that our proposed approach provides competitive performance for weakly supervised localization.

Jérôme Rony, Soufiane Belharbi, Jose Dolz et al.

Using deep learning models to diagnose cancer from histology data presents several challenges. Cancer grading and localization of regions of interest (ROIs) in these images normally relies on both image- and pixel-level labels, the latter requiring a costly annotation process. Deep weakly-supervised object localization (WSOL) methods provide different strategies for low-cost training of deep learning models. Using only image-class annotations, these methods can be trained to classify an image, and yield class activation maps (CAMs) for ROI localization. This paper provides a review of state-of-art DL methods for WSOL. We propose a taxonomy where these methods are divided into bottom-up and top-down methods according to the information flow in models. Although the latter have seen limited progress, recent bottom-up methods are currently driving much progress with deep WSOL methods. Early works focused on designing different spatial pooling functions. However, these methods reached limited localization accuracy, and unveiled a major limitation -- the under-activation of CAMs which leads to high false negative localization. Subsequent works aimed to alleviate this issue and recover complete object. Representative methods from our taxonomy are evaluated and compared in terms of classification and localization accuracy on two challenging histology datasets. Overall, the results indicate poor localization performance, particularly for generic methods that were initially designed to process natural images. Methods designed to address the challenges of histology data yielded good results. However, all methods suffer from high false positive/negative localization. Four key challenges are identified for the application of deep WSOL methods in histology -- under/over activation of CAMs, sensitivity to thresholding, and model selection.

Soufiane Belharbi

Neural network models and deep models are one of the leading and state of the art models in machine learning. Most successful deep neural models are the ones with many layers which highly increases their number of parameters. Training such models requires a large number of training samples which is not always available. One of the fundamental issues in neural networks is overfitting which is the issue tackled in this thesis. Such problem often occurs when the training of large models is performed using few training samples. Many approaches have been proposed to prevent the network from overfitting and improve its generalization performance such as data augmentation, early stopping, parameters sharing, unsupervised learning, dropout, batch normalization, etc. In this thesis, we tackle the neural network overfitting issue from a representation learning perspective by considering the situation where few training samples are available which is the case of many real world applications. We propose three contributions. The first one presented in chapter 2 is dedicated to dealing with structured output problems to perform multivariate regression when the output variable y contains structural dependencies between its components. The second contribution described in chapter 3 deals with the classification task where we propose to exploit prior knowledge about the internal representation of the hidden layers in neural networks. Our last contribution presented in chapter 4 showed the interest of transfer learning in applications where only few samples are available. In this contribution, we provide an automatic system based on such learning scheme with an application to medical domain. In this application, the task consists in localizing the third lumbar vertebra in a 3D CT scan. This work has been done in collaboration with the clinic Rouen Henri Becquerel Center who provided us with data.