Andrei Bursuc

Ioannis Kakogeorgiou, Spyros Gidaris, Bill Psomas et al.

Transformers and masked language modeling are quickly being adopted and explored in computer vision as vision transformers and masked image modeling (MIM). In this work, we argue that image token masking differs from token masking in text, due to the amount and correlation of tokens in an image. In particular, to generate a challenging pretext task for MIM, we advocate a shift from random masking to informed masking. We develop and exhibit this idea in the context of distillation-based MIM, where a teacher transformer encoder generates an attention map, which we use to guide masking for the student. We thus introduce a novel masking strategy, called attention-guided masking (AttMask), and we demonstrate its effectiveness over random masking for dense distillation-based MIM as well as plain distillation-based self-supervised learning on classification tokens. We confirm that AttMask accelerates the learning process and improves the performance on a variety of downstream tasks. We provide the implementation code at https://github.com/gkakogeorgiou/attmask.

Huy V. Vo, Oriane Siméoni, Spyros Gidaris et al.

Object detectors trained with weak annotations are affordable alternatives to fully-supervised counterparts. However, there is still a significant performance gap between them. We propose to narrow this gap by fine-tuning a base pre-trained weakly-supervised detector with a few fully-annotated samples automatically selected from the training set using ``box-in-box'' (BiB), a novel active learning strategy designed specifically to address the well-documented failure modes of weakly-supervised detectors. Experiments on the VOC07 and COCO benchmarks show that BiB outperforms other active learning techniques and significantly improves the base weakly-supervised detector's performance with only a few fully-annotated images per class. BiB reaches 97% of the performance of fully-supervised Fast RCNN with only 10% of fully-annotated images on VOC07. On COCO, using on average 10 fully-annotated images per class, or equivalently 1% of the training set, BiB also reduces the performance gap (in AP) between the weakly-supervised detector and the fully-supervised Fast RCNN by over 70%, showing a good trade-off between performance and data efficiency. Our code is publicly available at https://github.com/huyvvo/BiB.

Florent Bartoccioni, Éloi Zablocki, Andrei Bursuc et al.

Recent works in autonomous driving have widely adopted the bird's-eye-view (BEV) semantic map as an intermediate representation of the world. Online prediction of these BEV maps involves non-trivial operations such as multi-camera data extraction as well as fusion and projection into a common topview grid. This is usually done with error-prone geometric operations (e.g., homography or back-projection from monocular depth estimation) or expensive direct dense mapping between image pixels and pixels in BEV (e.g., with MLP or attention). In this work, we present 'LaRa', an efficient encoder-decoder, transformer-based model for vehicle semantic segmentation from multiple cameras. Our approach uses a system of cross-attention to aggregate information over multiple sensors into a compact, yet rich, collection of latent representations. These latent representations, after being processed by a series of self-attention blocks, are then reprojected with a second cross-attention in the BEV space. We demonstrate that our model outperforms the best previous works using transformers on nuScenes. The code and trained models are available at https://github.com/valeoai/LaRa

Angelika Ando, Spyros Gidaris, Andrei Bursuc et al.

Casting semantic segmentation of outdoor LiDAR point clouds as a 2D problem, e.g., via range projection, is an effective and popular approach. These projection-based methods usually benefit from fast computations and, when combined with techniques which use other point cloud representations, achieve state-of-the-art results. Today, projection-based methods leverage 2D CNNs but recent advances in computer vision show that vision transformers (ViTs) have achieved state-of-the-art results in many image-based benchmarks. In this work, we question if projection-based methods for 3D semantic segmentation can benefit from these latest improvements on ViTs. We answer positively but only after combining them with three key ingredients: (a) ViTs are notoriously hard to train and require a lot of training data to learn powerful representations. By preserving the same backbone architecture as for RGB images, we can exploit the knowledge from long training on large image collections that are much cheaper to acquire and annotate than point clouds. We reach our best results with pre-trained ViTs on large image datasets. (b) We compensate ViTs' lack of inductive bias by substituting a tailored convolutional stem for the classical linear embedding layer. (c) We refine pixel-wise predictions with a convolutional decoder and a skip connection from the convolutional stem to combine low-level but fine-grained features of the the convolutional stem with the high-level but coarse predictions of the ViT encoder. With these ingredients, we show that our method, called RangeViT, outperforms existing projection-based methods on nuScenes and SemanticKITTI. The code is available at https://github.com/valeoai/rangevit.

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.

Spyros Gidaris, Andrei Bursuc, Oriane Simeoni et al.

Self-supervised learning can be used for mitigating the greedy needs of Vision Transformer networks for very large fully-annotated datasets. Different classes of self-supervised learning offer representations with either good contextual reasoning properties, e.g., using masked image modeling strategies, or invariance to image perturbations, e.g., with contrastive methods. In this work, we propose a single-stage and standalone method, MOCA, which unifies both desired properties using novel mask-and-predict objectives defined with high-level features (instead of pixel-level details). Moreover, we show how to effectively employ both learning paradigms in a synergistic and computation-efficient way. Doing so, we achieve new state-of-the-art results on low-shot settings and strong experimental results in various evaluation protocols with a training that is at least 3 times faster than prior methods. We provide the implementation code at https://github.com/valeoai/MOCA.

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc et al.

Domain adaptation has been vastly investigated in computer vision but still requires access to target images at train time, which might be intractable in some uncommon conditions. In this paper, we propose the task of `Prompt-driven Zero-shot Domain Adaptation', where we adapt a model trained on a source domain using only a general description in natural language of the target domain, i.e., a prompt. First, we leverage a pretrained contrastive vision-language model (CLIP) to optimize affine transformations of source features, steering them towards the target text embedding while preserving their content and semantics. To achieve this, we propose Prompt-driven Instance Normalization (PIN). Second, we show that these prompt-driven augmentations can be used to perform zero-shot domain adaptation for semantic segmentation. Experiments demonstrate that our method significantly outperforms CLIP-based style transfer baselines on several datasets for the downstream task at hand, even surpassing one-shot unsupervised domain adaptation. A similar boost is observed on object detection and image classification. The code is available at https://github.com/astra-vision/PODA .

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc et al.

Generalization to new domains not seen during training is one of the long-standing challenges in deploying neural networks in real-world applications. Existing generalization techniques either necessitate external images for augmentation, and/or aim at learning invariant representations by imposing various alignment constraints. Large-scale pretraining has recently shown promising generalization capabilities, along with the potential of binding different modalities. For instance, the advent of vision-language models like CLIP has opened the doorway for vision models to exploit the textual modality. In this paper, we introduce a simple framework for generalizing semantic segmentation networks by employing language as the source of randomization. Our recipe comprises three key ingredients: (i) the preservation of the intrinsic CLIP robustness through minimal fine-tuning, (ii) language-driven local style augmentation, and (iii) randomization by locally mixing the source and augmented styles during training. Extensive experiments report state-of-the-art results on various generalization benchmarks. Code is accessible at https://github.com/astra-vision/FAMix .

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

Sebastian Cavada, Soumava Paul, Tuan-Hung Vu et al.

Previous work has evaluated physics reasoning in foundation models using synthetic or semi-synthetic scenes and visual question-answering tasks. However, these benchmarks emphasize high-level events and lack the visual fidelity required to assess true low-level Newtonian understanding. We introduce NewtPhys, a 4D physically annotated dataset built from multiview images of real-world scenes with physics-grounded simulations. The dataset provides dense, fine-grained annotations across timesteps -- including 3D forces and amodal per-pixel quantities covering physics, tracking, semantics and geometry -- bridging the gap between simplistic synthetic setups and realistic visual complexity. Using NewtPhys, we systematically evaluate 56 VLMs, including 54 open-weight models and 2 closed-source frontier models, and 10 VFMs and reveal limitations in low-level physics reasoning. Beyond benchmarking, our dataset enables future research in physics-grounded vision and the development of next-generation physics-aware evaluations. Code and datasets are available at https://astra-vision.github.io/NewtPhys.

Antonin Vobecky, David Hurych, Oriane Siméoni et al.

This work investigates learning pixel-wise semantic image segmentation in urban scenes without any manual annotation, just from the raw non-curated data collected by cars which, equipped with cameras and LiDAR sensors, drive around a city. Our contributions are threefold. First, we propose a novel method for cross-modal unsupervised learning of semantic image segmentation by leveraging synchronized LiDAR and image data. The key ingredient of our method is the use of an object proposal module that analyzes the LiDAR point cloud to obtain proposals for spatially consistent objects. Second, we show that these 3D object proposals can be aligned with the input images and reliably clustered into semantically meaningful pseudo-classes. Finally, we develop a cross-modal distillation approach that leverages image data partially annotated with the resulting pseudo-classes to train a transformer-based model for image semantic segmentation. We show the generalization capabilities of our method by testing on four different testing datasets (Cityscapes, Dark Zurich, Nighttime Driving and ACDC) without any finetuning, and demonstrate significant improvements compared to the current state of the art on this problem. See project webpage https://vobecant.github.io/DriveAndSegment/ for the code and more.

Sophia Sirko-Galouchenko, Monika Wysoczanska, Andrei Bursuc et al.

Multimodal large language models (MLLMs) perform well on many vision-language tasks but often struggle with vision-centric problems that require fine-grained visual reasoning. Recent evidence suggests that this limitation arises not from weak visual representations, but from under-utilization of visual information during instruction tuning, where many tasks can be partially solved using language priors alone. We propose a simple and lightweight approach that augments visual instruction tuning with a small number of visually grounded self-supervised tasks expressed as natural language instructions. By reformulating classical self-supervised pretext tasks, such as rotation prediction, color matching, and cross-view correspondence, as image-instruction-response triplets, we introduce supervision that cannot be solved without relying on visual evidence. Our approach requires no human annotations, no architectural modifications, and no additional training stages. Across multiple models, training regimes, and benchmarks, injecting only a small fraction (3-10%) of such visually grounded instructions consistently improves performance on vision-centric evaluations. Our findings highlight instruction tuning with visually grounded SSL tasks as a powerful lever for improving visual reasoning in MLLMs through simple adjustments to the training data distribution. Code available at: https://github.com/sirkosophia/V-GIFT

Gilles Puy, Spyros Gidaris, Alexandre Boulch et al.

Self-supervised image backbones can be used to address complex 2D tasks (e.g., semantic segmentation, object discovery) very efficiently and with little or no downstream supervision. Ideally, 3D backbones for lidar should be able to inherit these properties after distillation of these powerful 2D features. The most recent methods for image-to-lidar distillation on autonomous driving data show promising results, obtained thanks to distillation methods that keep improving. Yet, we still notice a large performance gap when measuring the quality of distilled and fully supervised features by linear probing. In this work, instead of focusing only on the distillation method, we study the effect of three pillars for distillation: the 3D backbone, the pretrained 2D backbones, and the pretraining dataset. In particular, thanks to our scalable distillation method named ScaLR, we show that scaling the 2D and 3D backbones and pretraining on diverse datasets leads to a substantial improvement of the feature quality. This allows us to significantly reduce the gap between the quality of distilled and fully-supervised 3D features, and to improve the robustness of the pretrained backbones to domain gaps and perturbations.

Subhrajyoti Dasgupta, Arindam Das, Senthil Yogamani et al.

Shadows are frequently encountered natural phenomena that significantly hinder the performance of computer vision perception systems in practical settings, e.g., autonomous driving. A solution to this would be to eliminate shadow regions from the images before the processing of the perception system. Yet, training such a solution requires pairs of aligned shadowed and non-shadowed images which are difficult to obtain. We introduce a novel weakly supervised shadow removal framework UnShadowNet trained using contrastive learning. It is composed of a DeShadower network responsible for the removal of the extracted shadow under the guidance of an Illumination network which is trained adversarially by the illumination critic and a Refinement network to further remove artefacts. We show that UnShadowNet can be easily extended to a fully-supervised set-up to exploit the ground-truth when available. UnShadowNet outperforms existing state-of-the-art approaches on three publicly available shadow datasets (ISTD, adjusted ISTD, SRD) in both the weakly and fully supervised setups.

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/.

Timo Sämann, Ahmed Mostafa Hammam, Andrei Bursuc et al.

Averaging predictions of a deep ensemble of networks is apopular and effective method to improve predictive performance andcalibration in various benchmarks and Kaggle competitions. However, theruntime and training cost of deep ensembles grow linearly with the size ofthe ensemble, making them unsuitable for many applications. Averagingensemble weights instead of predictions circumvents this disadvantageduring inference and is typically applied to intermediate checkpoints ofa model to reduce training cost. Albeit effective, only few works haveimproved the understanding and the performance of weight averaging.Here, we revisit this approach and show that a simple weight fusion (WF)strategy can lead to a significantly improved predictive performance andcalibration. We describe what prerequisites the weights must meet interms of weight space, functional space and loss. Furthermore, we presenta new test method (called oracle test) to measure the functional spacebetween weights. We demonstrate the versatility of our WF strategy acrossstate of the art segmentation CNNs and Transformers as well as real worlddatasets such as BDD100K and Cityscapes. We compare WF with similarapproaches and show our superiority for in- and out-of-distribution datain terms of predictive performance and calibration.

Tuan-Hung Vu, Eduardo Valle, Andrei Bursuc et al.

We propose the unified BRAVO challenge to benchmark the reliability of semantic segmentation models under realistic perturbations and unknown out-of-distribution (OOD) scenarios. We define two categories of reliability: (1) semantic reliability, which reflects the model's accuracy and calibration when exposed to various perturbations; and (2) OOD reliability, which measures the model's ability to detect object classes that are unknown during training. The challenge attracted nearly 100 submissions from international teams representing notable research institutions. The results reveal interesting insights into the importance of large-scale pre-training and minimal architectural design in developing robust and reliable semantic segmentation models.

Simon de Moreau, Yasser Almehio, Andrei Bursuc et al.

Nighttime camera-based depth estimation is a highly challenging task, especially for autonomous driving applications, where accurate depth perception is essential for ensuring safe navigation. Models trained on daytime data often fail in the absence of precise but costly LiDAR. Even vision foundation models trained on large amounts of data are unreliable in low-light conditions. In this work, we aim to improve the reliability of perception systems at night time. To this end, we introduce Light Enhanced Depth (LED), a novel, cost-effective approach that significantly improves depth estimation in low-light environments by harnessing a pattern projected by high definition headlights available in modern vehicles. LED leads to significant performance boosts across multiple depth-estimation architectures (encoder-decoder, Adabins, DepthFormer, Depth Anything V2) both on synthetic and real datasets. Furthermore, increased performances beyond illuminated areas reveal a holistic enhancement in scene understanding. Finally, we release the Nighttime Synthetic Drive Dataset, a synthetic and photo-realistic nighttime dataset, which comprises 49,990 comprehensively annotated images.

Monika Wysoczańska, Antonin Vobecky, Amaia Cardiel et al.

Recent CLIP-like Vision-Language Models (VLMs), pre-trained on large amounts of image-text pairs to align both modalities with a simple contrastive objective, have paved the way to open-vocabulary semantic segmentation. Given an arbitrary set of textual queries, image pixels are assigned the closest query in feature space. However, this works well when a user exhaustively lists all possible visual concepts in an image that contrast against each other for the assignment. This corresponds to the current evaluation setup in the literature, which relies on having access to a list of in-domain relevant concepts, typically classes of a benchmark dataset. Here, we consider the more challenging (and realistic) scenario of segmenting a single concept, given a textual prompt and nothing else. To achieve good results, besides contrasting with the generic 'background' text, we propose two different approaches to automatically generate, at test time, query-specific textual contrastive concepts. We do so by leveraging the distribution of text in the VLM's training set or crafted LLM prompts. We also propose a metric designed to evaluate this scenario and show the relevance of our approach on commonly used datasets.

Kevin Osanlou, Jeremy Frank, Andrei Bursuc et al.

Planning under uncertainty is an area of interest in artificial intelligence. We present a novel approach based on tree search and graph machine learning for the scheduling problem known as Disjunctive Temporal Networks with Uncertainty (DTNU). Dynamic Controllability (DC) of DTNUs seeks a reactive scheduling strategy to satisfy temporal constraints in response to uncontrollable action durations. We introduce new semantics for reactive scheduling: Time-based Dynamic Controllability (TDC) and a restricted subset of TDC, R-TDC. We design a tree search algorithm to determine whether or not a DTNU is R-TDC. Moreover, we leverage a graph neural network as a heuristic for tree search guidance. Finally, we conduct experiments on a known benchmark on which we show R-TDC to retain significant completeness with regard to DC, while being faster to prove. This results in the tree search processing fifty percent more DTNU problems in R-TDC than the state-of-the-art DC solver does in DC with the same time budget. We also observe that graph neural network search guidance leads to substantial performance gains on benchmarks of more complex DTNUs, with up to eleven times more problems solved than the baseline tree search.

Walter Simoncini, Spyros Gidaris, Andrei Bursuc et al.

This paper introduces FUNGI, Features from UNsupervised GradIents, a method to enhance the features of transformer encoders by leveraging self-supervised gradients. Our method is simple: given any pretrained model, we first compute gradients from various self-supervised objectives for each input. These gradients are projected to a lower dimension and then concatenated with the model's output embedding. The resulting features are evaluated on k-nearest neighbor classification over 11 datasets from vision, 5 from natural language processing, and 2 from audio. Across backbones spanning various sizes and pretraining strategies, FUNGI features provide consistent performance improvements over the embeddings. We also show that using FUNGI features can benefit linear classification, clustering and image retrieval, and that they significantly improve the retrieval-based in-context scene understanding abilities of pretrained models, for example improving upon DINO by +17% for semantic segmentation - without any training.

Ellington Kirby, Alexandre Boulch, Yihong Xu et al.

We present DrivoR, a simple and efficient transformer-based architecture for end-to-end autonomous driving. Our approach builds on pretrained Vision Transformers (ViTs) and introduces camera-aware register tokens that compress multi-camera features into a compact scene representation, significantly reducing downstream computation without sacrificing accuracy. These tokens drive two lightweight transformer decoders that generate and then score candidate trajectories. The scoring decoder learns to mimic an oracle and predicts interpretable sub-scores representing aspects such as safety, comfort, and efficiency, enabling behavior-conditioned driving at inference. Despite its minimal design, DrivoR outperforms or matches strong contemporary baselines across NAVSIM-v1, NAVSIM-v2, and the photorealistic closed-loop HUGSIM benchmark. Our results show that a pure-transformer architecture, combined with targeted token compression, is sufficient for accurate, efficient, and adaptive end-to-end driving. Code and checkpoints will be made available via the project page.

Victor Besnier, Andrei Bursuc, David Picard et al.

Recent works on predictive uncertainty estimation have shown promising results on Out-Of-Distribution (OOD) detection for semantic segmentation. However, these methods struggle to precisely locate the point of interest in the image, i.e, the anomaly. This limitation is due to the difficulty of finegrained prediction at the pixel level. To address this issue, we build upon the recent ObsNet approach by providing object instance knowledge to the observer. We extend ObsNet by harnessing an instance-wise mask prediction. We use an additional, class agnostic, object detector to filter and aggregate observer predictions. Finally, we predict an unique anomaly score for each instance in the image. We show that our proposed method accurately disentangles in-distribution objects from OOD objects on three datasets.

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.

Monika Wysoczańska, Oriane Siméoni, Michaël Ramamonjisoa et al.

The popular CLIP model displays impressive zero-shot capabilities thanks to its seamless interaction with arbitrary text prompts. However, its lack of spatial awareness makes it unsuitable for dense computer vision tasks, e.g., semantic segmentation, without an additional fine-tuning step that often uses annotations and can potentially suppress its original open-vocabulary properties. Meanwhile, self-supervised representation methods have demonstrated good localization properties without human-made annotations nor explicit supervision. In this work, we take the best of both worlds and propose an open-vocabulary semantic segmentation method, which does not require any annotations. We propose to locally improve dense MaskCLIP features, which are computed with a simple modification of CLIP's last pooling layer, by integrating localization priors extracted from self-supervised features. By doing so, we greatly improve the performance of MaskCLIP and produce smooth outputs. Moreover, we show that the used self-supervised feature properties can directly be learnt from CLIP features. Our method CLIP-DINOiser needs only a single forward pass of CLIP and two light convolutional layers at inference, no extra supervision nor extra memory and reaches state-of-the-art results on challenging and fine-grained benchmarks such as COCO, Pascal Context, Cityscapes and ADE20k. The code to reproduce our results is available at https://github.com/wysoczanska/clip_dinoiser.

Ryousuke Yamada, Kohsuke Ide, Yoshihiro Fukuhara et al.

Despite recent progress in 3D self-supervised learning, collecting large-scale 3D scene scans remains expensive and labor-intensive. In this work, we investigate whether 3D representations can be learned from unlabeled videos recorded without any real 3D sensors. We present Laplacian-Aware Multi-level 3D Clustering with Sinkhorn-Knopp (LAM3C), a self-supervised framework that learns from video-generated point clouds from unlabeled videos. We first introduce RoomTours, a video-generated point cloud dataset constructed by collecting room-walkthrough videos from the web (e.g., real-estate tours) and generating 49,219 scenes using an off-the-shelf feed-forward reconstruction model. We also propose a noise-regularized loss that stabilizes representation learning by enforcing local geometric smoothness and ensuring feature stability under noisy point clouds. Remarkably, without using any real 3D scans, LAM3C achieves higher performance than the previous self-supervised methods on indoor semantic and instance segmentation. These results suggest that unlabeled videos represent an abundant source of data for 3D self-supervised learning.

Sophia Sirko-Galouchenko, Alexandre Boulch, Spyros Gidaris et al.

We introduce a self-supervised pretraining method, called OccFeat, for camera-only Bird's-Eye-View (BEV) segmentation networks. With OccFeat, we pretrain a BEV network via occupancy prediction and feature distillation tasks. Occupancy prediction provides a 3D geometric understanding of the scene to the model. However, the geometry learned is class-agnostic. Hence, we add semantic information to the model in the 3D space through distillation from a self-supervised pretrained image foundation model. Models pretrained with our method exhibit improved BEV semantic segmentation performance, particularly in low-data scenarios. Moreover, empirical results affirm the efficacy of integrating feature distillation with 3D occupancy prediction in our pretraining approach. Repository: https://github.com/valeoai/Occfeat

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.

Florent Bartoccioni, Elias Ramzi, Victor Besnier et al.

We explore the potential of large-scale generative video models for autonomous driving, introducing an open-source auto-regressive video model (VaViM) and its companion video-action model (VaVAM) to investigate how video pre-training transfers to real-world driving. VaViM is a simple auto-regressive video model that predicts frames using spatio-temporal token sequences. We show that it captures the semantics and dynamics of driving scenes. VaVAM, the video-action model, leverages the learned representations of VaViM to generate driving trajectories through imitation learning. Together, the models form a complete perception-to-action pipeline. We evaluate our models in open- and closed-loop driving scenarios, revealing that video-based pre-training holds promise for autonomous driving. Key insights include the semantic richness of the learned representations, the benefits of scaling for video synthesis, and the complex relationship between model size, data, and safety metrics in closed-loop evaluations. We release code and model weights at https://github.com/valeoai/VideoActionModel

Shashanka Venkataramanan, Valentinos Pariza, Mohammadreza Salehi et al.

We present Franca (pronounced Fran-ka): free one; the first fully open-source (data, code, weights) vision foundation model that matches and in many cases surpasses the performance of state-of-the-art proprietary models, e.g., DINOv2, CLIP, SigLIPv2, etc. Our approach is grounded in a transparent training pipeline inspired by Web-SSL and uses publicly available data: ImageNet-21K and a subset of ReLAION-2B. Beyond model release, we tackle critical limitations in SSL clustering methods. While modern models rely on assigning image features to large codebooks via clustering algorithms like Sinkhorn-Knopp, they fail to account for the inherent ambiguity in clustering semantics. To address this, we introduce a parameter-efficient, multi-head clustering projector based on nested Matryoshka representations. This design progressively refines features into increasingly fine-grained clusters without increasing the model size, enabling both performance and memory efficiency. Additionally, we propose a novel positional disentanglement strategy that explicitly removes positional biases from dense representations, thereby improving the encoding of semantic content. This leads to consistent gains on several downstream benchmarks, demonstrating the utility of cleaner feature spaces. Our contributions establish a new standard for transparent, high-performance vision models and open a path toward more reproducible and generalizable foundation models for the broader AI community. The code and model checkpoints are available at https://github.com/valeoai/Franca.

Yasser Benigmim, Mohammad Fahes, Tuan-Hung Vu et al.

In this paper, we challenge the conventional practice in Open-Vocabulary Semantic Segmentation (OVSS) of using averaged class-wise text embeddings, which are typically obtained by encoding each class name with multiple templates (e.g., a photo of <class>, a sketch of a <class>). We investigate the impact of templates for OVSS, and find that for each class, there exist single-template classifiers--which we refer to as class-experts--that significantly outperform the conventional averaged classifier. First, to identify these class-experts, we introduce a novel approach that estimates them without any labeled data or training. By leveraging the class-wise prediction entropy of single-template classifiers, we select those yielding the lowest entropy as the most reliable class-experts. Second, we combine the outputs of class-experts in a new fusion process. Our plug-and-play method, coined FLOSS, is orthogonal and complementary to existing OVSS methods, offering an improvement without the need for additional labels or training. Extensive experiments show that FLOSS consistently enhances state-of-the-art OVSS models, generalizes well across datasets with different distribution shifts, and delivers substantial improvements in low-data scenarios where only a few unlabeled images are available. Our code is available at https://github.com/yasserben/FLOSS .

Sophia Sirko-Galouchenko, Spyros Gidaris, Antonin Vobecky et al.

We introduce DIP, a novel unsupervised post-training method designed to enhance dense image representations in large-scale pretrained vision encoders for in-context scene understanding. Unlike prior approaches that rely on complex self-distillation architectures, our method trains the vision encoder using pseudo-tasks that explicitly simulate downstream in-context scenarios, inspired by meta-learning principles. To enable post-training on unlabeled data, we propose an automatic mechanism for generating in-context tasks that combines a pretrained diffusion model and the vision encoder itself. DIP is simple, unsupervised, and computationally efficient, requiring less than 9 hours on a single A100 GPU. By learning dense representations through pseudo in-context tasks, it achieves strong performance across a wide variety of downstream real-world in-context scene understanding tasks. It outperforms both the initial vision encoder and prior methods, offering a practical and effective solution for improving dense representations. Code available here: https://github.com/sirkosophia/DIP

Yihong Xu, Éloi Zablocki, Alexandre Boulch et al.

Motion forecasting is crucial in autonomous driving systems to anticipate the future trajectories of surrounding agents such as pedestrians, vehicles, and traffic signals. In end-to-end forecasting, the model must jointly detect and track from sensor data (cameras or LiDARs) the past trajectories of the different elements of the scene and predict their future locations. We depart from the current trend of tackling this task via end-to-end training from perception to forecasting, and instead use a modular approach. We individually build and train detection, tracking and forecasting modules. We then only use consecutive finetuning steps to integrate the modules better and alleviate compounding errors. We conduct an in-depth study on the finetuning strategies and it reveals that our simple yet effective approach significantly improves performance on the end-to-end forecasting benchmark. Consequently, our solution ranks first in the Argoverse 2 End-to-end Forecasting Challenge, with 63.82 mAPf. We surpass forecasting results by +17.1 points over last year's winner and by +13.3 points over this year's runner-up. This remarkable performance in forecasting can be explained by our modular paradigm, which integrates finetuning strategies and significantly outperforms the end-to-end-trained counterparts. The code, model weights and results are made available https://github.com/valeoai/valeo4cast.

Oriane Siméoni, Gilles Puy, Huy V. Vo et al.

Localizing objects in image collections without supervision can help to avoid expensive annotation campaigns. We propose a simple approach to this problem, that leverages the activation features of a vision transformer pre-trained in a self-supervised manner. Our method, LOST, does not require any external object proposal nor any exploration of the image collection; it operates on a single image. Yet, we outperform state-of-the-art object discovery methods by up to 8 CorLoc points on PASCAL VOC 2012. We also show that training a class-agnostic detector on the discovered objects boosts results by another 7 points. Moreover, we show promising results on the unsupervised object discovery task. The code to reproduce our results can be found at https://github.com/valeoai/LOST.

Spyros Gidaris, Andrei Bursuc, Gilles Puy et al.

Learning image representations without human supervision is an important and active research field. Several recent approaches have successfully leveraged the idea of making such a representation invariant under different types of perturbations, especially via contrastive-based instance discrimination training. Although effective visual representations should indeed exhibit such invariances, there are other important characteristics, such as encoding contextual reasoning skills, for which alternative reconstruction-based approaches might be better suited. With this in mind, we propose a teacher-student scheme to learn representations by training a convolutional net to reconstruct a bag-of-visual-words (BoW) representation of an image, given as input a perturbed version of that same image. Our strategy performs an online training of both the teacher network (whose role is to generate the BoW targets) and the student network (whose role is to learn representations), along with an online update of the visual-words vocabulary (used for the BoW targets). This idea effectively enables fully online BoW-guided unsupervised learning. Extensive experiments demonstrate the interest of our BoW-based strategy which surpasses previous state-of-the-art methods (including contrastive-based ones) in several applications. For instance, in downstream tasks such Pascal object detection, Pascal classification and Places205 classification, our method improves over all prior unsupervised approaches, thus establishing new state-of-the-art results that are also significantly better even than those of supervised pre-training. We provide the implementation code at https://github.com/valeoai/obow.

Antonin Vobecky, Oriane Siméoni, David Hurych et al.

We describe an approach to predict open-vocabulary 3D semantic voxel occupancy map from input 2D images with the objective of enabling 3D grounding, segmentation and retrieval of free-form language queries. This is a challenging problem because of the 2D-3D ambiguity and the open-vocabulary nature of the target tasks, where obtaining annotated training data in 3D is difficult. The contributions of this work are three-fold. First, we design a new model architecture for open-vocabulary 3D semantic occupancy prediction. The architecture consists of a 2D-3D encoder together with occupancy prediction and 3D-language heads. The output is a dense voxel map of 3D grounded language embeddings enabling a range of open-vocabulary tasks. Second, we develop a tri-modal self-supervised learning algorithm that leverages three modalities: (i) images, (ii) language and (iii) LiDAR point clouds, and enables training the proposed architecture using a strong pre-trained vision-language model without the need for any 3D manual language annotations. Finally, we demonstrate quantitatively the strengths of the proposed model on several open-vocabulary tasks: Zero-shot 3D semantic segmentation using existing datasets; 3D grounding and retrieval of free-form language queries, using a small dataset that we propose as an extension of nuScenes. You can find the project page here https://vobecant.github.io/POP3D.

Simon de Moreau, Andrei Bursuc, Hafid El-Idrissi et al.

Nighttime environments pose significant challenges for camera-based perception, as existing methods passively rely on the scene lighting. We introduce Lighting-driven Dynamic Active Sensing (LiDAS), a closed-loop active illumination system that combines off-the-shelf visual perception models with high-definition headlights. Rather than uniformly brightening the scene, LiDAS dynamically predicts an optimal illumination field that maximizes downstream perception performance, i.e., decreasing light on empty areas to reallocate it on object regions. LiDAS enables zero-shot nighttime generalization of daytime-trained models through adaptive illumination control. Trained on synthetic data and deployed zero-shot in real-world closed-loop driving scenarios, LiDAS enables +18.7% mAP50 and +5.0% mIoU over standard low-beam at equal power. It maintains performances while reducing energy use by 40%. LiDAS complements domain-generalization methods, further strengthening robustness without retraining. By turning readily available headlights into active vision actuators, LiDAS offers a cost-effective solution to robust nighttime perception.

Simon de Moreau, Mathias Corsia, Hassan Bouchiba et al.

Accurate depth information is essential for many computer vision applications. Yet, no available dataset recording method allows for fully dense accurate depth estimation in a large scale dynamic environment. In this paper, we introduce DOC-Depth, a novel, efficient and easy-to-deploy approach for dense depth generation from any LiDAR sensor. After reconstructing consistent dense 3D environment using LiDAR odometry, we address dynamic objects occlusions automatically thanks to DOC, our state-of-the art dynamic object classification method. Additionally, DOC-Depth is fast and scalable, allowing for the creation of unbounded datasets in terms of size and time. We demonstrate the effectiveness of our approach on the KITTI dataset, improving its density from 16.1% to 71.2% and release this new fully dense depth annotation, to facilitate future research in the domain. We also showcase results using various LiDAR sensors and in multiple environments. All software components are publicly available for the research community.

Victor Letzelter, Hugo Malard, Mathieu Fontaine et al.

We propose LoRA-MCL, a training scheme that extends next-token prediction in language models with a method designed to decode diverse, plausible sentence continuations at inference time. Traditional language modeling is an intrinsically ill-posed problem: given a context, multiple futures may be equally plausible. Our approach leverages Multiple Choice Learning (MCL) and the Winner-Takes-All (WTA) loss to efficiently handle ambiguity through Low-Rank Adaptation (LoRA). We provide a theoretical interpretation of applying Multiple Choice Learning to Language Modeling, assuming the data is generated from a mixture of distributions. To illustrate the proposed approach, we use data sampled from mixtures of Markov chains. We then demonstrate with extensive experiments on real-world visual and audio captioning tasks that our method achieves high diversity and relevance in generated outputs.

Victor Besnier, David Hurych, Andrei Bursuc et al.

Recent advances in image and video generation have raised significant interest from both academia and industry. A key challenge in this field is improving inference efficiency, as model size and the number of inference steps directly impact the commercial viability of generative models while also posing fundamental scientific challenges. A promising direction involves combining auto-regressive sequential token modeling with multi-token prediction per step, reducing inference time by up to an order of magnitude. However, predicting multiple tokens in parallel can introduce structural inconsistencies due to token incompatibilities, as capturing complex joint dependencies during training remains challenging. Traditionally, once tokens are sampled, there is no mechanism to backtrack and refine erroneous predictions. We propose a method for self-correcting image generation by iteratively refining sampled tokens. We achieve this with a novel training scheme that injects random tokens in the context, improving robustness and enabling token fixing during sampling. Our method preserves the efficiency benefits of parallel token prediction while significantly enhancing generation quality. We evaluate our approach on image generation using the ImageNet-256 and CIFAR-10 datasets, as well as on video generation with UCF-101 and NuScenes, demonstrating substantial improvements across both modalities.

Yuan Yin, Shashanka Venkataramanan, Tuan-Hung Vu et al.

Parameter-efficient fine-tuning (PEFT) methods, such as LoRA, reduce adaptation cost by injecting low-rank updates into pretrained weights. However, LoRA's down-projection is randomly initialized and data-agnostic, discarding potentially useful information. Prior analyses show that this projection changes little during training, while the up-projection carries most of the adaptation, making the random input compression a performance bottleneck. We propose IPA, a feature-aware projection framework that explicitly preserves information in the reduced hidden space. In the linear case, we instantiate IPA with algorithms approximating top principal components, enabling efficient projector pretraining with negligible inference overhead. Across language and vision benchmarks, IPA consistently improves over LoRA and DoRA, achieving on average 1.5 points higher accuracy on commonsense reasoning and 2.3 points on VTAB-1k, while matching full LoRA performance with roughly half the trainable parameters when the projection is frozen.

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc et al.

Domain adaptation has been extensively investigated in computer vision but still requires access to target data at the training time, which might be difficult to obtain in some uncommon conditions. In this paper, we present a new framework for domain adaptation relying on a single Vision-Language (VL) latent embedding instead of full target data. First, leveraging a contrastive language-image pre-training model (CLIP), we propose prompt/photo-driven instance normalization (PIN). PIN is a feature augmentation method that mines multiple visual styles using a single target VL latent embedding, by optimizing affine transformations of low-level source features. The VL embedding can come from a language prompt describing the target domain, a partially optimized language prompt, or a single unlabeled target image. Second, we show that these mined styles (i.e., augmentations) can be used for zero-shot (i.e., target-free) and one-shot unsupervised domain adaptation. Experiments on semantic segmentation demonstrate the effectiveness of the proposed method, which outperforms relevant baselines in the zero-shot and one-shot settings.

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.

Corentin Sautier, Gilles Puy, Spyros Gidaris et al.

Segmenting or detecting objects in sparse Lidar point clouds are two important tasks in autonomous driving to allow a vehicle to act safely in its 3D environment. The best performing methods in 3D semantic segmentation or object detection rely on a large amount of annotated data. Yet annotating 3D Lidar data for these tasks is tedious and costly. In this context, we propose a self-supervised pre-training method for 3D perception models that is tailored to autonomous driving data. Specifically, we leverage the availability of synchronized and calibrated image and Lidar sensors in autonomous driving setups for distilling self-supervised pre-trained image representations into 3D models. Hence, our method does not require any point cloud nor image annotations. The key ingredient of our method is the use of superpixels which are used to pool 3D point features and 2D pixel features in visually similar regions. We then train a 3D network on the self-supervised task of matching these pooled point features with the corresponding pooled image pixel features. The advantages of contrasting regions obtained by superpixels are that: (1) grouping together pixels and points of visually coherent regions leads to a more meaningful contrastive task that produces features well adapted to 3D semantic segmentation and 3D object detection; (2) all the different regions have the same weight in the contrastive loss regardless of the number of 3D points sampled in these regions; (3) it mitigates the noise produced by incorrect matching of points and pixels due to occlusions between the different sensors. Extensive experiments on autonomous driving datasets demonstrate the ability of our image-to-Lidar distillation strategy to produce 3D representations that transfer well on semantic segmentation and object detection tasks.

Victor Besnier, Andrei Bursuc, David Picard et al.

In this paper, we tackle the detection of out-of-distribution (OOD) objects in semantic segmentation. By analyzing the literature, we found that current methods are either accurate or fast but not both which limits their usability in real world applications. To get the best of both aspects, we propose to mitigate the common shortcomings by following four design principles: decoupling the OOD detection from the segmentation task, observing the entire segmentation network instead of just its output, generating training data for the OOD detector by leveraging blind spots in the segmentation network and focusing the generated data on localized regions in the image to simulate OOD objects. Our main contribution is a new OOD detection architecture called ObsNet associated with a dedicated training scheme based on Local Adversarial Attacks (LAA). We validate the soundness of our approach across numerous ablation studies. We also show it obtains top performances both in speed and accuracy when compared to ten recent methods of the literature on three different datasets.

Kevin Osanlou, Jeremy Frank, J. Benton et al.

Scheduling in the presence of uncertainty is an area of interest in artificial intelligence due to the large number of applications. We study the problem of dynamic controllability (DC) of disjunctive temporal networks with uncertainty (DTNU), which seeks a strategy to satisfy all constraints in response to uncontrollable action durations. We introduce a more restricted, stronger form of controllability than DC for DTNUs, time-based dynamic controllability (TDC), and present a tree search approach to determine whether or not a DTNU is TDC. Moreover, we leverage the learning capability of a message passing neural network (MPNN) as a heuristic for tree search guidance. Finally, we conduct experiments for which the tree search shows superior results to state-of-the-art timed-game automata (TGA) based approaches. We observe that using an MPNN for tree search guidance leads to a significant increase in solving performance and scalability to harder DTNU problems.

Kevin Osanlou, Christophe Guettier, Andrei Bursuc et al.

Learning-based methods are increasingly popular for search algorithms in single-criterion optimization problems. In contrast, for multiple-criteria optimization there are significantly fewer approaches despite the existence of numerous applications. Constrained path-planning for Autonomous Ground Vehicles (AGV) is one such application, where an AGV is typically deployed in disaster relief or search and rescue applications in off-road environments. The agent can be faced with the following dilemma : optimize a source-destination path according to a known criterion and an uncertain criterion under operational constraints. The known criterion is associated to the cost of the path, representing the distance. The uncertain criterion represents the feasibility of driving through the path without requiring human intervention. It depends on various external parameters such as the physics of the vehicle, the state of the explored terrains or weather conditions. In this work, we leverage knowledge acquired through offline simulations by training a neural network model to predict the uncertain criterion. We integrate this model inside a path-planner which can solve problems online. Finally, we conduct experiments on realistic AGV scenarios which illustrate that the proposed framework requires human intervention less frequently, trading for a limited increase in the path distance.

Kevin Osanlou, Andrei Bursuc, Christophe Guettier et al.

Deep learning-based methods are growing prominence for planning purposes. In this paper, we present a hybrid planner that combines a graph machine learning model and an optimal solver based on branch and bound tree search for path-planning tasks. More specifically, a graph neural network is used to assist the branch and bound algorithm in handling constraints associated with a desired solution path. There are multiple downstream practical applications, such as Autonomous Unmanned Ground Vehicles (AUGV), typically deployed in disaster relief or search and rescue operations. In off-road environments, AUGVs must dynamically optimize a source-destination path under various operational constraints, out of which several are difficult to predict in advance and need to be addressed online. We conduct experiments on realistic scenarios and show that graph neural network support enables substantial speedup and smoother scaling to harder path-planning problems. Additionally, information provided by the graph neural network enables the approach to outperform problem-specific handcrafted heuristics, highlighting the potential graph neural networks hold for path-planning tasks.