Patrick Pérez

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

Mehdi Zemni, Mickaël Chen, Éloi Zablocki et al.

Nowadays, deep vision models are being widely deployed in safety-critical applications, e.g., autonomous driving, and explainability of such models is becoming a pressing concern. Among explanation methods, counterfactual explanations aim to find minimal and interpretable changes to the input image that would also change the output of the model to be explained. Such explanations point end-users at the main factors that impact the decision of the model. However, previous methods struggle to explain decision models trained on images with many objects, e.g., urban scenes, which are more difficult to work with but also arguably more critical to explain. In this work, we propose to tackle this issue with an object-centric framework for counterfactual explanation generation. Our method, inspired by recent generative modeling works, encodes the query image into a latent space that is structured in a way to ease object-level manipulations. Doing so, it provides the end-user with control over which search directions (e.g., spatial displacement of objects, style modification, etc.) are to be explored during the counterfactual generation. We conduct a set of experiments on counterfactual explanation benchmarks for driving scenes, and we show that our method can be adapted beyond classification, e.g., to explain semantic segmentation models. To complete our analysis, we design and run a user study that measures the usefulness of counterfactual explanations in understanding a decision model. Code is available at https://github.com/valeoai/OCTET.

Yihong Xu, Loïck Chambon, Éloi Zablocki et al.

Motion forecasting is crucial in enabling autonomous vehicles to anticipate the future trajectories of surrounding agents. To do so, it requires solving mapping, detection, tracking, and then forecasting problems, in a multi-step pipeline. In this complex system, advances in conventional forecasting methods have been made using curated data, i.e., with the assumption of perfect maps, detection, and tracking. This paradigm, however, ignores any errors from upstream modules. Meanwhile, an emerging end-to-end paradigm, that tightly integrates the perception and forecasting architectures into joint training, promises to solve this issue. However, the evaluation protocols between the two methods were so far incompatible and their comparison was not possible. In fact, conventional forecasting methods are usually not trained nor tested in real-world pipelines (e.g., with upstream detection, tracking, and mapping modules). In this work, we aim to bring forecasting models closer to the real-world deployment. First, we propose a unified evaluation pipeline for forecasting methods with real-world perception inputs, allowing us to compare conventional and end-to-end methods for the first time. Second, our in-depth study uncovers a substantial performance gap when transitioning from curated to perception-based data. In particular, we show that this gap (1) stems not only from differences in precision but also from the nature of imperfect inputs provided by perception modules, and that (2) is not trivially reduced by simply finetuning on perception outputs. Based on extensive experiments, we provide recommendations for critical areas that require improvement and guidance towards more robust motion forecasting in the real world. The evaluation library for benchmarking models under standardized and practical conditions is provided: \url{https://github.com/valeoai/MFEval}.

Oriane Siméoni, Chloé Sekkat, Gilles Puy et al.

Recent advances in self-supervised visual representation learning have paved the way for unsupervised methods tackling tasks such as object discovery and instance segmentation. However, discovering objects in an image with no supervision is a very hard task; what are the desired objects, when to separate them into parts, how many are there, and of what classes? The answers to these questions depend on the tasks and datasets of evaluation. In this work, we take a different approach and propose to look for the background instead. This way, the salient objects emerge as a by-product without any strong assumption on what an object should be. We propose FOUND, a simple model made of a single $conv1\times1$ initialized with coarse background masks extracted from self-supervised patch-based representations. After fast training and refining these seed masks, the model reaches state-of-the-art results on unsupervised saliency detection and object discovery benchmarks. Moreover, we show that our approach yields good results in the unsupervised semantic segmentation retrieval task. The code to reproduce our results is available at https://github.com/valeoai/FOUND.

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 .

Alexandre Défossez, Laurent Mazaré, Manu Orsini et al.

We introduce Moshi, a speech-text foundation model and full-duplex spoken dialogue framework. Current systems for spoken dialogue rely on pipelines of independent components, namely voice activity detection, speech recognition, textual dialogue and text-to-speech. Such frameworks cannot emulate the experience of real conversations. First, their complexity induces a latency of several seconds between interactions. Second, text being the intermediate modality for dialogue, non-linguistic information that modifies meaning -- such as emotion or non-speech sounds -- is lost in the interaction. Finally, they rely on a segmentation into speaker turns, which does not take into account overlapping speech, interruptions and interjections. Moshi solves these independent issues altogether by casting spoken dialogue as speech-to-speech generation. Starting from a text language model backbone, Moshi generates speech as tokens from the residual quantizer of a neural audio codec, while modeling separately its own speech and that of the user into parallel streams. This allows for the removal of explicit speaker turns, and the modeling of arbitrary conversational dynamics. We moreover extend the hierarchical semantic-to-acoustic token generation of previous work to first predict time-aligned text tokens as a prefix to audio tokens. Not only this "Inner Monologue" method significantly improves the linguistic quality of generated speech, but we also illustrate how it can provide streaming speech recognition and text-to-speech. Our resulting model is the first real-time full-duplex spoken large language model, with a theoretical latency of 160ms, 200ms in practice, and is available at https://github.com/kyutai-labs/moshi.

Oriane Siméoni, Éloi Zablocki, Spyros Gidaris et al.

The recent enthusiasm for open-world vision systems show the high interest of the community to perform perception tasks outside of the closed-vocabulary benchmark setups which have been so popular until now. Being able to discover objects in images/videos without knowing in advance what objects populate the dataset is an exciting prospect. But how to find objects without knowing anything about them? Recent works show that it is possible to perform class-agnostic unsupervised object localization by exploiting self-supervised pre-trained features. We propose here a survey of unsupervised object localization methods that discover objects in images without requiring any manual annotation in the era of self-supervised ViTs. We gather links of discussed methods in the repository https://github.com/valeoai/Awesome-Unsupervised-Object-Localization.

Soshi Shimada, Vladislav Golyanik, Zhi Li et al.

Marker-less monocular 3D human motion capture (MoCap) with scene interactions is a challenging research topic relevant for extended reality, robotics and virtual avatar generation. Due to the inherent depth ambiguity of monocular settings, 3D motions captured with existing methods often contain severe artefacts such as incorrect body-scene inter-penetrations, jitter and body floating. To tackle these issues, we propose HULC, a new approach for 3D human MoCap which is aware of the scene geometry. HULC estimates 3D poses and dense body-environment surface contacts for improved 3D localisations, as well as the absolute scale of the subject. Furthermore, we introduce a 3D pose trajectory optimisation based on a novel pose manifold sampling that resolves erroneous body-environment inter-penetrations. Although the proposed method requires less structured inputs compared to existing scene-aware monocular MoCap algorithms, it produces more physically-plausible poses: HULC significantly and consistently outperforms the existing approaches in various experiments and on different metrics. Project page: https://vcai.mpi-inf.mpg.de/projects/HULC/.

Awet Haileslassie Gebrehiwot, David Hurych, Karel Zimmermann et al.

Deep perception models have to reliably cope with an open-world setting of domain shifts induced by different geographic regions, sensor properties, mounting positions, and several other reasons. Since covering all domains with annotated data is technically intractable due to the endless possible variations, researchers focus on unsupervised domain adaptation (UDA) methods that adapt models trained on one (source) domain with annotations available to another (target) domain for which only unannotated data are available. Current predominant methods either leverage semi-supervised approaches, e.g., teacher-student setup, or exploit privileged data, such as other sensor modalities or temporal data consistency. We introduce a novel domain adaptation method that leverages the best of both trends. Our approach combines input data's temporal and cross-sensor geometric consistency with the mean teacher method. Dubbed T-UDA for "temporal UDA", such a combination yields massive performance gains for the task of 3D semantic segmentation of driving scenes. Experiments are conducted on Waymo Open Dataset, nuScenes and SemanticKITTI, for two popular 3D point cloud architectures, Cylinder3D and MinkowskiNet. Our codes are publicly available at https://github.com/ctu-vras/T-UDA.

Soshi Shimada, Vladislav Golyanik, Patrick Pérez et al.

Existing methods for 3D tracking from monocular RGB videos predominantly consider articulated and rigid objects. Modelling dense non-rigid object deformations in this setting remained largely unaddressed so far, although such effects can improve the realism of the downstream applications such as AR/VR and avatar communications. This is due to the severe ill-posedness of the monocular view setting and the associated challenges. While it is possible to naively track multiple non-rigid objects independently using 3D templates or parametric 3D models, such an approach would suffer from multiple artefacts in the resulting 3D estimates such as depth ambiguity, unnatural intra-object collisions and missing or implausible deformations. Hence, this paper introduces the first method that addresses the fundamental challenges depicted above and that allows tracking human hands interacting with human faces in 3D from single monocular RGB videos. We model hands as articulated objects inducing non-rigid face deformations during an active interaction. Our method relies on a new hand-face motion and interaction capture dataset with realistic face deformations acquired with a markerless multi-view camera system. As a pivotal step in its creation, we process the reconstructed raw 3D shapes with position-based dynamics and an approach for non-uniform stiffness estimation of the head tissues, which results in plausible annotations of the surface deformations, hand-face contact regions and head-hand positions. At the core of our neural approach are a variational auto-encoder supplying the hand-face depth prior and modules that guide the 3D tracking by estimating the contacts and the deformations. Our final 3D hand and face reconstructions are realistic and more plausible compared to several baselines applicable in our setting, both quantitatively and qualitatively. https://vcai.mpi-inf.mpg.de/projects/Decaf

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.

Antoine Saporta, Arthur Douillard, Tuan-Hung Vu et al.

Unsupervised Domain Adaptation (UDA) is a transfer learning task which aims at training on an unlabeled target domain by leveraging a labeled source domain. Beyond the traditional scope of UDA with a single source domain and a single target domain, real-world perception systems face a variety of scenarios to handle, from varying lighting conditions to many cities around the world. In this context, UDAs with several domains increase the challenges with the addition of distribution shifts within the different target domains. This work focuses on a novel framework for learning UDA, continuous UDA, in which models operate on multiple target domains discovered sequentially, without access to previous target domains. We propose MuHDi, for Multi-Head Distillation, a method that solves the catastrophic forgetting problem, inherent in continual learning tasks. MuHDi performs distillation at multiple levels from the previous model as well as an auxiliary target-specialist segmentation head. We report both extensive ablation and experiments on challenging multi-target UDA semantic segmentation benchmarks to validate the proposed learning scheme and architecture.

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.

Laura Calem, Hedi Ben-Younes, Patrick Pérez et al.

Predicting multiple trajectories for road users is important for automated driving systems: ego-vehicle motion planning indeed requires a clear view of the possible motions of the surrounding agents. However, the generative models used for multiple-trajectory forecasting suffer from a lack of diversity in their proposals. To avoid this form of collapse, we propose a novel method for structured prediction of diverse trajectories. To this end, we complement an underlying pretrained generative model with a diversity component, based on a determinantal point process (DPP). We balance and structure this diversity with the inclusion of knowledge-based quality constraints, independent from the underlying generative model. We combine these two novel components with a gating operation, ensuring that the predictions are both diverse and within the drivable area. We demonstrate on the nuScenes driving dataset the relevance of our compound approach, which yields significant improvements in the diversity and the quality of the generated trajectories.

Léon Zheng, Gilles Puy, Elisa Riccietti et al.

We introduce a regularization loss based on kernel mean embeddings with rotation-invariant kernels on the hypersphere (also known as dot-product kernels) for self-supervised learning of image representations. Besides being fully competitive with the state of the art, our method significantly reduces time and memory complexity for self-supervised training, making it implementable for very large embedding dimensions on existing devices and more easily adjustable than previous methods to settings with limited resources. Our work follows the major paradigm where the model learns to be invariant to some predefined image transformations (cropping, blurring, color jittering, etc.), while avoiding a degenerate solution by regularizing the embedding distribution. Our particular contribution is to propose a loss family promoting the embedding distribution to be close to the uniform distribution on the hypersphere, with respect to the maximum mean discrepancy pseudometric. We demonstrate that this family encompasses several regularizers of former methods, including uniformity-based and information-maximization methods, which are variants of our flexible regularization loss with different kernels. Beyond its practical consequences for state-of-the-art self-supervised learning with limited resources, the proposed generic regularization approach opens perspectives to leverage more widely the literature on kernel methods in order to improve self-supervised learning methods.

Victor Letzelter, Mathieu Fontaine, Mickaël Chen et al.

We introduce Resilient Multiple Choice Learning (rMCL), an extension of the MCL approach for conditional distribution estimation in regression settings where multiple targets may be sampled for each training input. Multiple Choice Learning is a simple framework to tackle multimodal density estimation, using the Winner-Takes-All (WTA) loss for a set of hypotheses. In regression settings, the existing MCL variants focus on merging the hypotheses, thereby eventually sacrificing the diversity of the predictions. In contrast, our method relies on a novel learned scoring scheme underpinned by a mathematical framework based on Voronoi tessellations of the output space, from which we can derive a probabilistic interpretation. After empirically validating rMCL with experiments on synthetic data, we further assess its merits on the sound source localization problem, demonstrating its practical usefulness and the relevance of its interpretation.

Simon Roburin, Charles Corbière, Gilles Puy et al.

Predictive performance of machine learning models trained with empirical risk minimization (ERM) can degrade considerably under distribution shifts. The presence of spurious correlations in training datasets leads ERM-trained models to display high loss when evaluated on minority groups not presenting such correlations. Extensive attempts have been made to develop methods improving worst-group robustness. However, they require group information for each training input or at least, a validation set with group labels to tune their hyperparameters, which may be expensive to get or unknown a priori. In this paper, we address the challenge of improving group robustness without group annotation during training or validation. To this end, we propose to partition the training dataset into groups based on Gram matrices of features extracted by an ``identification'' model and to apply robust optimization based on these pseudo-groups. In the realistic context where no group labels are available, our experiments show that our approach not only improves group robustness over ERM but also outperforms all recent baselines

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.

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

Neil Zeghidour, Eugene Kharitonov, Manu Orsini et al.

We introduce Delayed Streams Modeling (DSM), a flexible formulation for streaming, multimodal sequence-to-sequence learning. Sequence-to-sequence generation is often cast in an offline manner, where the model consumes the complete input sequence before generating the first output timestep. Alternatively, streaming sequence-to-sequence rely on learning a policy for choosing when to advance on the input stream, or write to the output stream. DSM instead models already time-aligned streams with a decoder-only language model. By moving the alignment to a pre-processing step,and introducing appropriate delays between streams, DSM provides streaming inference of arbitrary output sequences, from any input combination, making it applicable to many sequence-to-sequence problems. In particular, given text and audio streams, automatic speech recognition (ASR) corresponds to the text stream being delayed, while the opposite gives a text-to-speech (TTS) model. We perform extensive experiments for these two major sequence-to-sequence tasks, showing that DSM provides state-of-the-art performance and latency while supporting arbitrary long sequences, being even competitive with offline baselines. Code, samples and demos are available at https://github.com/kyutai-labs/delayed-streams-modeling

Hippolyte Pilchen, Edouard Grave, Patrick Pérez

Recent techniques such as retrieval-augmented generation or chain-of-thought reasoning have led to longer contexts and increased inference costs. Context compression techniques can reduce these costs, but the most effective approaches require fine-tuning the target model or even modifying its architecture. This can degrade its general abilities when not used for this specific purpose. Here we explore an alternative approach: an encoder that compresses the context into continuous representations which replace token embeddings in decoder LLMs. First, we perform a systematic study of training strategies and architecture choices for the encoder. Our findings led to the design of an Adaptable text Representations Compressor, named ARC-Encoder, which outputs $x$-times fewer continuous representations (typically $x\!\in\!\{4,8\}$) than text tokens. We evaluate ARC-Encoder across a variety of LLM usage scenarios, ranging from in-context learning to context window extension, on both instruct and base decoders. Results show that ARC-Encoder achieves state-of-the-art performance on several benchmarks while improving computational efficiency at inference. Finally, we demonstrate that our models can be adapted to multiple decoders simultaneously, allowing a single encoder to generalize across different decoder LLMs. This makes ARC-Encoder a flexible and efficient solution for portable encoders that work seamlessly with multiple LLMs. We release a training code at https://github.com/kyutai-labs/ARC-Encoder , fine-tuning dataset and pretrained models are available at https://huggingface.co/collections/kyutai/arc-encoders-68ee18787301407d60a57047 .

Julien Rebut, Arthur Ouaknine, Waqas Malik et al.

With their robustness to adverse weather conditions and ability to measure speeds, radar sensors have been part of the automotive landscape for more than two decades. Recent progress toward High Definition (HD) Imaging radar has driven the angular resolution below the degree, thus approaching laser scanning performance. However, the amount of data a HD radar delivers and the computational cost to estimate the angular positions remain a challenge. In this paper, we propose a novel HD radar sensing model, FFT-RadNet, that eliminates the overhead of computing the range-azimuth-Doppler 3D tensor, learning instead to recover angles from a range-Doppler spectrum. FFT-RadNet is trained both to detect vehicles and to segment free driving space. On both tasks, it competes with the most recent radar-based models while requiring less compute and memory. Also, we collected and annotated 2-hour worth of raw data from synchronized automotive-grade sensors (camera, laser, HD radar) in various environments (city street, highway, countryside road). This unique dataset, nick-named RADIal for "Radar, Lidar et al.", is available at https://github.com/valeoai/RADIal.

Paul Jacob, Éloi Zablocki, Hédi Ben-Younes et al.

As deep learning models are increasingly used in safety-critical applications, explainability and trustworthiness become major concerns. For simple images, such as low-resolution face portraits, synthesizing visual counterfactual explanations has recently been proposed as a way to uncover the decision mechanisms of a trained classification model. In this work, we address the problem of producing counterfactual explanations for high-quality images and complex scenes. Leveraging recent semantic-to-image models, we propose a new generative counterfactual explanation framework that produces plausible and sparse modifications which preserve the overall scene structure. Furthermore, we introduce the concept of "region-targeted counterfactual explanations", and a corresponding framework, where users can guide the generation of counterfactuals by specifying a set of semantic regions of the query image the explanation must be about. Extensive experiments are conducted on challenging datasets including high-quality portraits (CelebAMask-HQ) and driving scenes (BDD100k). Code is available at https://github.com/valeoai/STEEX

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.

Hédi Ben-Younes, Éloi Zablocki, Mickaël Chen et al.

Learning-based trajectory prediction models have encountered great success, with the promise of leveraging contextual information in addition to motion history. Yet, we find that state-of-the-art forecasting methods tend to overly rely on the agent's current dynamics, failing to exploit the semantic contextual cues provided at its input. To alleviate this issue, we introduce CAB, a motion forecasting model equipped with a training procedure designed to promote the use of semantic contextual information. We also introduce two novel metrics - dispersion and convergence-to-range - to measure the temporal consistency of successive forecasts, which we found missing in standard metrics. Our method is evaluated on the widely adopted nuScenes Prediction benchmark as well as on a subset of the most difficult examples from this benchmark. The code is available at github.com/valeoai/CAB

Huy V. Vo, Elena Sizikova, Cordelia Schmid et al.

Existing approaches to unsupervised object discovery (UOD) do not scale up to large datasets without approximations that compromise their performance. We propose a novel formulation of UOD as a ranking problem, amenable to the arsenal of distributed methods available for eigenvalue problems and link analysis. Through the use of self-supervised features, we also demonstrate the first effective fully unsupervised pipeline for UOD. Extensive experiments on COCO and OpenImages show that, in the single-object discovery setting where a single prominent object is sought in each image, the proposed LOD (Large-scale Object Discovery) approach is on par with, or better than the state of the art for medium-scale datasets (up to 120K images), and over 37% better than the only other algorithms capable of scaling up to 1.7M images. In the multi-object discovery setting where multiple objects are sought in each image, the proposed LOD is over 14% better in average precision (AP) than all other methods for datasets ranging from 20K to 1.7M images. Using self-supervised features, we also show that the proposed method obtains state-of-the-art UOD performance on OpenImages. Our code is publicly available at https://github.com/huyvvo/LOD.

Arthur Ouaknine, Alasdair Newson, Patrick Pérez et al.

Understanding the scene around the ego-vehicle is key to assisted and autonomous driving. Nowadays, this is mostly conducted using cameras and laser scanners, despite their reduced performances in adverse weather conditions. Automotive radars are low-cost active sensors that measure properties of surrounding objects, including their relative speed, and have the key advantage of not being impacted by rain, snow or fog. However, they are seldom used for scene understanding due to the size and complexity of radar raw data and the lack of annotated datasets. Fortunately, recent open-sourced datasets have opened up research on classification, object detection and semantic segmentation with raw radar signals using end-to-end trainable models. In this work, we propose several novel architectures, and their associated losses, which analyse multiple "views" of the range-angle-Doppler radar tensor to segment it semantically. Experiments conducted on the recent CARRADA dataset demonstrate that our best model outperforms alternative models, derived either from the semantic segmentation of natural images or from radar scene understanding, while requiring significantly fewer parameters. Both our code and trained models are available at https://github.com/valeoai/MVRSS.

Maximilian Jaritz, Tuan-Hung Vu, Raoul de Charette et al.

Domain adaptation is an important task to enable learning when labels are scarce. While most works focus only on the image modality, there are many important multi-modal datasets. In order to leverage multi-modality for domain adaptation, we propose cross-modal learning, where we enforce consistency between the predictions of two modalities via mutual mimicking. We constrain our network to make correct predictions on labeled data and consistent predictions across modalities on unlabeled target-domain data. Experiments in unsupervised and semi-supervised domain adaptation settings prove the effectiveness of this novel domain adaptation strategy. Specifically, we evaluate on the task of 3D semantic segmentation from either the 2D image, the 3D point cloud or from both. We leverage recent driving datasets to produce a wide variety of domain adaptation scenarios including changes in scene layout, lighting, sensor setup and weather, as well as the synthetic-to-real setup. Our method significantly improves over previous uni-modal adaptation baselines on all adaption scenarios. Our code is publicly available at https://github.com/valeoai/xmuda_journal

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.

Adithya Ranga, Filippo Giruzzi, Jagdish Bhanushali et al.

Advanced perception and path planning are at the core for any self-driving vehicle. Autonomous vehicles need to understand the scene and intentions of other road users for safe motion planning. For urban use cases it is very important to perceive and predict the intentions of pedestrians, cyclists, scooters, etc., classified as vulnerable road users (VRU). Intent is a combination of pedestrian activities and long term trajectories defining their future motion. In this paper we propose a multi-task learning model to predict pedestrian actions, crossing intent and forecast their future path from video sequences. We have trained the model on naturalistic driving open-source JAAD dataset, which is rich in behavioral annotations and real world scenarios. Experimental results show state-of-the-art performance on JAAD dataset and how we can benefit from jointly learning and predicting actions and trajectories using 2D human pose features and scene context.

Maxime Bucher, Tuan-Hung Vu, Matthieu Cord et al.

In this work, we define and address a novel domain adaptation (DA) problem in semantic scene segmentation, where the target domain not only exhibits a data distribution shift w.r.t. the source domain, but also includes novel classes that do not exist in the latter. Different to "open-set" and "universal domain adaptation", which both regard all objects from new classes as "unknown", we aim at explicit test-time prediction for these new classes. To reach this goal, we propose a framework that leverages domain adaptation and zero-shot learning techniques to enable "boundless" adaptation in the target domain. It relies on a novel architecture, along with a dedicated learning scheme, to bridge the source-target domain gap while learning how to map new classes' labels to relevant visual representations. The performance is further improved using self-training on target-domain pseudo-labels. For validation, we consider different domain adaptation set-ups, namely synthetic-2-real, country-2-country and dataset-2-dataset. Our framework outperforms the baselines by significant margins, setting competitive standards on all benchmarks for the new task. Code and models are available at https://github.com/valeoai/buda.

Maximilian Jaritz, Tuan-Hung Vu, Raoul de Charette et al.

Unsupervised Domain Adaptation (UDA) is crucial to tackle the lack of annotations in a new domain. There are many multi-modal datasets, but most UDA approaches are uni-modal. In this work, we explore how to learn from multi-modality and propose cross-modal UDA (xMUDA) where we assume the presence of 2D images and 3D point clouds for 3D semantic segmentation. This is challenging as the two input spaces are heterogeneous and can be impacted differently by domain shift. In xMUDA, modalities learn from each other through mutual mimicking, disentangled from the segmentation objective, to prevent the stronger modality from adopting false predictions from the weaker one. We evaluate on new UDA scenarios including day-to-night, country-to-country and dataset-to-dataset, leveraging recent autonomous driving datasets. xMUDA brings large improvements over uni-modal UDA on all tested scenarios, and is complementary to state-of-the-art UDA techniques. Code is available at https://github.com/valeoai/xmuda.

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.

Moritz Böhle, Amélie Royer, Juliette Marrie et al.

Vision-language models (VLMs) are commonly trained by inserting image tokens from a pretrained vision encoder into the textual stream of a language model. This allows text and image information to fully attend to one another within the model, but becomes extremely costly for high-resolution images, long conversations, or streaming videos, both in memory and compute. VLMs leveraging cross-attention are an efficient alternative to token insertion but exhibit a clear performance gap, in particular on tasks involving fine-grained visual details. We find that a key to improving such models is to also enable local text-to-text interaction in the dedicated cross-attention layers. Building on this, we propose CASA, Cross-Attention via Self-Attention, a simple and efficient paradigm which substantially reduces the gap with full token insertion on common image understanding benchmarks, while enjoying the same scalability as cross-attention models when applied to long-context multimodal tasks such as streaming video captioning. For samples and code, please see our project page at https://kyutai.org/casa .

Thibaut Loiseau, Tuan-Hung Vu, Mickael Chen et al.

Assessing the robustness of perception models to covariate shifts and their ability to detect out-of-distribution (OOD) inputs is crucial for safety-critical applications such as autonomous vehicles. By nature of such applications, however, the relevant data is difficult to collect and annotate. In this paper, we show for the first time how synthetic data can be specifically generated to assess comprehensively the real-world reliability of semantic segmentation models. By fine-tuning Stable Diffusion with only in-domain data, we perform zero-shot generation of visual scenes in OOD domains or inpainted with OOD objects. This synthetic data is employed to evaluate the robustness of pretrained segmenters, thereby offering insights into their performance when confronted with real edge cases. Through extensive experiments, we demonstrate a high correlation between the performance of models when evaluated on our synthetic OOD data and when evaluated on real OOD inputs, showing the relevance of such virtual testing. Furthermore, we demonstrate how our approach can be utilized to enhance the calibration and OOD detection capabilities of segmenters. Code and data are made public.

Tom Labiausse, Laurent Mazaré, Edouard Grave et al.

We introduce Hibiki, a decoder-only model for simultaneous speech translation. Hibiki leverages a multistream language model to synchronously process source and target speech, and jointly produces text and audio tokens to perform speech-to-text and speech-to-speech translation. We furthermore address the fundamental challenge of simultaneous interpretation, which unlike its consecutive counterpart, where one waits for the end of the source utterance to start translating, adapts its flow to accumulate just enough context to produce a correct translation in real-time, chunk by chunk. To do so, we introduce a weakly-supervised method that leverages the perplexity of an off-the-shelf text translation system to identify optimal delays on a per-word basis and create aligned synthetic data. After supervised training, Hibiki performs adaptive, simultaneous speech translation with vanilla temperature sampling. On a French-English simultaneous speech translation task, Hibiki demonstrates state-of-the-art performance in translation quality, speaker fidelity and naturalness. Moreover, the simplicity of its inference process makes it compatible with batched translation and even real-time on-device deployment. We provide examples as well as models and inference code.

Cédric Rommel, Victor Letzelter, Nermin Samet et al.

We propose ManiPose, a manifold-constrained multi-hypothesis model for human-pose 2D-to-3D lifting. We provide theoretical and empirical evidence that, due to the depth ambiguity inherent to monocular 3D human pose estimation, traditional regression models suffer from pose-topology consistency issues, which standard evaluation metrics (MPJPE, P-MPJPE and PCK) fail to assess. ManiPose addresses depth ambiguity by proposing multiple candidate 3D poses for each 2D input, each with its estimated plausibility. Unlike previous multi-hypothesis approaches, ManiPose forgoes generative models, greatly facilitating its training and usage. By constraining the outputs to lie on the human pose manifold, ManiPose guarantees the consistency of all hypothetical poses, in contrast to previous works. We showcase the performance of ManiPose on real-world datasets, where it outperforms state-of-the-art models in pose consistency by a large margin while being very competitive on the MPJPE metric.

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.

Kaouther Messaoud, Kathrin Grosse, Mickael Chen et al.

Autonomous vehicles ought to predict the surrounding agents' trajectories to allow safe maneuvers in uncertain and complex traffic situations. As companies increasingly apply trajectory prediction in the real world, security becomes a relevant concern. In this paper, we focus on backdoors - a security threat acknowledged in other fields but so far overlooked for trajectory prediction. To this end, we describe and investigate four triggers that could affect trajectory prediction. We then show that these triggers (for example, a braking vehicle), when correlated with a desired output (for example, a curve) during training, cause the desired output of a state-of-the-art trajectory prediction model. In other words, the model has good benign performance but is vulnerable to backdoors. This is the case even if the trigger maneuver is performed by a non-casual agent behind the target vehicle. As a side-effect, our analysis reveals interesting limitations within trajectory prediction models. Finally, we evaluate a range of defenses against backdoors. While some, like simple offroad checks, do not enable detection for all triggers, clustering is a promising candidate to support manual inspection to find backdoors.

Amélie Royer, Moritz Böhle, Gabriel de Marmiesse et al.

The recent successes of Vision-Language models raise the question of how to equivalently imbue a pretrained speech model with vision understanding, an important milestone towards building a multimodal speech model able to freely converse about images. Building such a conversational Vision-Speech model brings its unique challenges: (i) paired image-speech datasets are much scarcer than their image-text counterparts, (ii) ensuring real-time latency at inference is crucial thus bringing compute and memory constraints, and (iii) the model should preserve prosodic features (e.g., speaker tone) which cannot be inferred from text alone. In this work, we introduce MoshiVis, augmenting a recent dialogue speech LLM, Moshi, with visual inputs through lightweight adaptation modules. An additional dynamic gating mechanism enables the model to more easily switch between the visual inputs and unrelated conversation topics. To reduce training costs, we design a simple one-stage, parameter-efficient fine-tuning pipeline in which we leverage a mixture of image-text (i.e., "speechless") and image-speech samples. We evaluate the model on downstream visual understanding tasks with both audio and text prompts, and report qualitative samples of interactions with MoshiVis. Our inference code will be made available, as well as the image-speech data used for audio evaluation.

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.

Victor Letzelter, David Perera, Cédric Rommel et al.

Winner-takes-all training is a simple learning paradigm, which handles ambiguous tasks by predicting a set of plausible hypotheses. Recently, a connection was established between Winner-takes-all training and centroidal Voronoi tessellations, showing that, once trained, hypotheses should quantize optimally the shape of the conditional distribution to predict. However, the best use of these hypotheses for uncertainty quantification is still an open question. In this work, we show how to leverage the appealing geometric properties of the Winner-takes-all learners for conditional density estimation, without modifying its original training scheme. We theoretically establish the advantages of our novel estimator both in terms of quantization and density estimation, and we demonstrate its competitiveness on synthetic and real-world datasets, including audio data.

Cédric Rommel, Eduardo Valle, Mickaël Chen et al.

We present an innovative approach to 3D Human Pose Estimation (3D-HPE) by integrating cutting-edge diffusion models, which have revolutionized diverse fields, but are relatively unexplored in 3D-HPE. We show that diffusion models enhance the accuracy, robustness, and coherence of human pose estimations. We introduce DiffHPE, a novel strategy for harnessing diffusion models in 3D-HPE, and demonstrate its ability to refine standard supervised 3D-HPE. We also show how diffusion models lead to more robust estimations in the face of occlusions, and improve the time-coherence and the sagittal symmetry of predictions. Using the Human\,3.6M dataset, we illustrate the effectiveness of our approach and its superiority over existing models, even under adverse situations where the occlusion patterns in training do not match those in inference. Our findings indicate that while standalone diffusion models provide commendable performance, their accuracy is even better in combination with supervised models, opening exciting new avenues for 3D-HPE research.

Himalaya Jain, Tuan-Hung Vu, Patrick Pérez et al.

With the rapid advances in generative adversarial networks (GANs), the visual quality of synthesised scenes keeps improving, including for complex urban scenes with applications to automated driving. We address in this work a continual scene generation setup in which GANs are trained on a stream of distinct domains; ideally, the learned models should eventually be able to generate new scenes in all seen domains. This setup reflects the real-life scenario where data are continuously acquired in different places at different times. In such a continual setup, we aim for learning with zero forgetting, \IE, with no degradation in synthesis quality over earlier domains due to catastrophic forgetting. To this end, we introduce a novel framework that not only (i) enables seamless knowledge transfer in continual training but also (ii) guarantees zero forgetting with a small overhead cost. While being more memory efficient, thanks to continual learning, our model obtains better synthesis quality as compared against the brute-force solution that trains one full model for each domain. Especially, under extreme low-data regimes, our approach outperforms the brute-force one by a large margin.

Florent Bartoccioni, Éloi Zablocki, Patrick Pérez et al.

Vision-based depth estimation is a key feature in autonomous systems, which often relies on a single camera or several independent ones. In such a monocular setup, dense depth is obtained with either additional input from one or several expensive LiDARs, e.g., with 64 beams, or camera-only methods, which suffer from scale-ambiguity and infinite-depth problems. In this paper, we propose a new alternative of densely estimating metric depth by combining a monocular camera with a light-weight LiDAR, e.g., with 4 beams, typical of today's automotive-grade mass-produced laser scanners. Inspired by recent self-supervised methods, we introduce a novel framework, called LiDARTouch, to estimate dense depth maps from monocular images with the help of ``touches'' of LiDAR, i.e., without the need for dense ground-truth depth. In our setup, the minimal LiDAR input contributes on three different levels: as an additional model's input, in a self-supervised LiDAR reconstruction objective function, and to estimate changes of pose (a key component of self-supervised depth estimation architectures). Our LiDARTouch framework achieves new state of the art in self-supervised depth estimation on the KITTI dataset, thus supporting our choices of integrating the very sparse LiDAR signal with other visual features. Moreover, we show that the use of a few-beam LiDAR alleviates scale ambiguity and infinite-depth issues that camera-only methods suffer from. We also demonstrate that methods from the fully-supervised depth-completion literature can be adapted to a self-supervised regime with a minimal LiDAR signal.

Antoine Saporta, Tuan-Hung Vu, Matthieu Cord et al.

In this work, we address the task of unsupervised domain adaptation (UDA) for semantic segmentation in presence of multiple target domains: The objective is to train a single model that can handle all these domains at test time. Such a multi-target adaptation is crucial for a variety of scenarios that real-world autonomous systems must handle. It is a challenging setup since one faces not only the domain gap between the labeled source set and the unlabeled target set, but also the distribution shifts existing within the latter among the different target domains. To this end, we introduce two adversarial frameworks: (i) multi-discriminator, which explicitly aligns each target domain to its counterparts, and (ii) multi-target knowledge transfer, which learns a target-agnostic model thanks to a multi-teacher/single-student distillation mechanism.The evaluation is done on four newly-proposed multi-target benchmarks for UDA in semantic segmentation. In all tested scenarios, our approaches consistently outperform baselines, setting competitive standards for the novel task.

Guillaume Le Moing, Tuan-Hung Vu, Himalaya Jain et al.

Despite the recent progress of generative adversarial networks (GANs) at synthesizing photo-realistic images, producing complex urban scenes remains a challenging problem. Previous works break down scene generation into two consecutive phases: unconditional semantic layout synthesis and image synthesis conditioned on layouts. In this work, we propose to condition layout generation as well for higher semantic control: given a vector of class proportions, we generate layouts with matching composition. To this end, we introduce a conditional framework with novel architecture designs and learning objectives, which effectively accommodates class proportions to guide the scene generation process. The proposed architecture also allows partial layout editing with interesting applications. Thanks to the semantic control, we can produce layouts close to the real distribution, helping enhance the whole scene generation process. On different metrics and urban scene benchmarks, our models outperform existing baselines. Moreover, we demonstrate the merit of our approach for data augmentation: semantic segmenters trained on real layout-image pairs along with additional ones generated by our approach outperform models only trained on real pairs.