Renaud Marlet

Tetiana Martyniuk, Jonathan Seele, Alexandre Boulch et al.

This paper investigates "free lunch" strategies to boost the performance of lidar semantic scene completion (SSC) without requiring complex architectural redesigns. We first demonstrate that endowing input point clouds with semantic pseudo-labels from off-the-shelf segmentors significantly improves the performance of existing architectures. By evaluating these models against an oracle, we establish that high-quality semantic priors are a primary driver of mIoU gains. Furthermore, we equip the input lidar scan with visibility information that distinguishes between empty and unknown spaces, which provides a secondary performance boost across the tested architectures. Using these simple enhancements, we observe that older models remain competitive with state-of-the-art systems, and can even outperform them. Our code is available at https://github.com/astra-vision/SSC-Priors.

Georgy Ponimatkin, Nermin Samet, Yang Xiao et al.

We propose a simple, yet powerful approach for unsupervised object segmentation in videos. We introduce an objective function whose minimum represents the mask of the main salient object over the input sequence. It only relies on independent image features and optical flows, which can be obtained using off-the-shelf self-supervised methods. It scales with the length of the sequence with no need for superpixels or sparsification, and it generalizes to different datasets without any specific training. This objective function can actually be derived from a form of spectral clustering applied to the entire video. Our method achieves on-par performance with the state of the art on standard benchmarks (DAVIS2016, SegTrack-v2, FBMS59), while being conceptually and practically much simpler. Code is available at https://ponimatkin.github.io/ssl-vos.

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.

Alexandre Boulch, Corentin Sautier, Björn Michele et al.

We propose a new self-supervised method for pre-training the backbone of deep perception models operating on point clouds. The core idea is to train the model on a pretext task which is the reconstruction of the surface on which the 3D points are sampled, and to use the underlying latent vectors as input to the perception head. The intuition is that if the network is able to reconstruct the scene surface, given only sparse input points, then it probably also captures some fragments of semantic information, that can be used to boost an actual perception task. This principle has a very simple formulation, which makes it both easy to implement and widely applicable to a large range of 3D sensors and deep networks performing semantic segmentation or object detection. In fact, it supports a single-stream pipeline, as opposed to most contrastive learning approaches, allowing training on limited resources. We conducted extensive experiments on various autonomous driving datasets, involving very different kinds of lidars, for both semantic segmentation and object detection. The results show the effectiveness of our method to learn useful representations without any annotation, compared to existing approaches. Code is available at https://github.com/valeoai/ALSO

Nermin Samet, Oriane Siméoni, Gilles Puy et al.

We propose SeedAL, a method to seed active learning for efficient annotation of 3D point clouds for semantic segmentation. Active Learning (AL) iteratively selects relevant data fractions to annotate within a given budget, but requires a first fraction of the dataset (a 'seed') to be already annotated to estimate the benefit of annotating other data fractions. We first show that the choice of the seed can significantly affect the performance of many AL methods. We then propose a method for automatically constructing a seed that will ensure good performance for AL. Assuming that images of the point clouds are available, which is common, our method relies on powerful unsupervised image features to measure the diversity of the point clouds. It selects the point clouds for the seed by optimizing the diversity under an annotation budget, which can be done by solving a linear optimization problem. Our experiments demonstrate the effectiveness of our approach compared to random seeding and existing methods on both the S3DIS and SemanticKitti datasets. Code is available at https://github.com/nerminsamet/seedal.

Raphael Sulzer, Renaud Marlet, Bruno Vallet et al.

We present a comprehensive survey and benchmark of both traditional and learning-based methods for surface reconstruction from point clouds. This task is particularly challenging for real-world acquisitions due to factors such as noise, outliers, non-uniform sampling, and missing data. Traditional approaches often simplify the problem by imposing handcrafted priors on either the input point clouds or the resulting surface, a process that can require tedious hyperparameter tuning. In contrast, deep learning models have the capability to directly learn the properties of input point clouds and desired surfaces from data. We study the influence of handcrafted and learned priors on the precision and robustness of surface reconstruction techniques. We evaluate various time-tested and contemporary methods in a standardized manner. When both trained and evaluated on point clouds with identical characteristics, the learning-based models consistently produce higher-quality surfaces compared to their traditional counterparts -- even in scenarios involving novel shape categories. However, traditional methods demonstrate greater resilience to the diverse anomalies commonly found in real-world 3D acquisitions. For the benefit of the research community, we make our code and datasets available, inviting further enhancements to learning-based surface reconstruction. This can be accessed at https://github.com/raphaelsulzer/dsr-benchmark .

Gilles Puy, Nermin Samet, Alexandre Boulch et al.

Plain Transformers have become the de-facto architecture for processing text, audio, image, and video, offering a unified backbone for multimodal learning. However, state-of-the-art architectures for point cloud semantic segmentation remain dominated by U-Nets architectures where convolutions are interleaved with local or windowed attentions. In this work, we show how to effectively leverage vanilla, non-hierarchical ViTs for segmentation of large-scale automotive lidar scenes. We bridge the performance gap thanks to a carefully designed tokenizer, a lightweight decoder segmentation head, and tailored data augmentations. Our approach, VaViT for Vanilla ViT, matches or exceeds the performance of state-of-the-art methods while maintaining the simplicity of ViT architecture. We provide extensive evaluations on nuScenes, SemanticKITTI, and Waymo Open Dataset to validate the efficiency of our method. Code and models are available at https://github.com/valeoai/VaViT.

Björn Michele, Alexandre Boulch, Tuan-Hung Vu et al.

We tackle the challenging problem of source-free unsupervised domain adaptation (SFUDA) for 3D semantic segmentation. It amounts to performing domain adaptation on an unlabeled target domain without any access to source data; the available information is a model trained to achieve good performance on the source domain. A common issue with existing SFUDA approaches is that performance degrades after some training time, which is a by product of an under-constrained and ill-posed problem. We discuss two strategies to alleviate this issue. First, we propose a sensible way to regularize the learning problem. Second, we introduce a novel criterion based on agreement with a reference model. It is used (1) to stop the training when appropriate and (2) as validator to select hyperparameters without any knowledge on the target domain. Our contributions are easy to implement and readily amenable for all SFUDA methods, ensuring stable improvements over all baselines. We validate our findings on various 3D lidar settings, achieving state-of-the-art performance. The project repository (with code) is: github.com/valeoai/TTYD.

Nermin Samet, Gilles Puy, Renaud Marlet

This paper presents a new method for the zero-shot open-vocabulary semantic segmentation (OVSS) of 3D automotive lidar data. To circumvent the recognized image-text modality gap that is intrinsic to approaches based on Vision Language Models (VLMs) such as CLIP, our method relies instead on image generation from text, to create prototype images. Given a 3D network distilled from a 2D Vision Foundation Model (VFM), we then label a point cloud by matching 3D point features with 2D image features of these prototypes. Our method is state-of-the-art for OVSS on nuScenes and SemanticKITTI. Code, pre-trained models, and generated images are available at https://github.com/valeoai/IGLOSS.

Gilles Puy, Alexandre Boulch, Renaud Marlet

Semantic segmentation of point clouds in autonomous driving datasets requires techniques that can process large numbers of points efficiently. Sparse 3D convolutions have become the de-facto tools to construct deep neural networks for this task: they exploit point cloud sparsity to reduce the memory and computational loads and are at the core of today's best methods. In this paper, we propose an alternative method that reaches the level of state-of-the-art methods without requiring sparse convolutions. We actually show that such level of performance is achievable by relying on tools a priori unfit for large scale and high-performing 3D perception. In particular, we propose a novel 3D backbone, WaffleIron, made almost exclusively of MLPs and dense 2D convolutions and present how to train it to reach high performance on SemanticKITTI and nuScenes. We believe that WaffleIron is a compelling alternative to backbones using sparse 3D convolutions, especially in frameworks and on hardware where those convolutions are not readily available.

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.

Björn Michele, Alexandre Boulch, Gilles Puy et al.

Learning models on one labeled dataset that generalize well on another domain is a difficult task, as several shifts might happen between the data domains. This is notably the case for lidar data, for which models can exhibit large performance discrepancies due for instance to different lidar patterns or changes in acquisition conditions. This paper addresses the corresponding Unsupervised Domain Adaptation (UDA) task for semantic segmentation. To mitigate this problem, we introduce an unsupervised auxiliary task of learning an implicit underlying surface representation simultaneously on source and target data. As both domains share the same latent representation, the model is forced to accommodate discrepancies between the two sources of data. This novel strategy differs from classical minimization of statistical divergences or lidar-specific domain adaptation techniques. Our experiments demonstrate that our method achieves a better performance than the current state of the art, both in real-to-real and synthetic-to-real scenarios.

Corentin Sautier, Gilles Puy, Alexandre Boulch et al.

We present a surprisingly simple and efficient method for self-supervision of 3D backbone on automotive Lidar point clouds. We design a contrastive loss between features of Lidar scans captured in the same scene. Several such approaches have been proposed in the literature from PointConstrast, which uses a contrast at the level of points, to the state-of-the-art TARL, which uses a contrast at the level of segments, roughly corresponding to objects. While the former enjoys a great simplicity of implementation, it is surpassed by the latter, which however requires a costly pre-processing. In BEVContrast, we define our contrast at the level of 2D cells in the Bird's Eye View plane. Resulting cell-level representations offer a good trade-off between the point-level representations exploited in PointContrast and segment-level representations exploited in TARL: we retain the simplicity of PointContrast (cell representations are cheap to compute) while surpassing the performance of TARL in downstream semantic segmentation.

Corentin Sautier, Gilles Puy, Alexandre Boulch et al.

Online object segmentation and tracking in Lidar point clouds enables autonomous agents to understand their surroundings and make safe decisions. Unfortunately, manual annotations for these tasks are prohibitively costly. We tackle this problem with the task of class-agnostic unsupervised online instance segmentation and tracking. To that end, we leverage an instance segmentation backbone and propose a new training recipe that enables the online tracking of objects. Our network is trained on pseudo-labels, eliminating the need for manual annotations. We conduct an evaluation using metrics adapted for temporal instance segmentation. Computing these metrics requires temporally-consistent instance labels. When unavailable, we construct these labels using the available 3D bounding boxes and semantic labels in the dataset. We compare our method against strong baselines and demonstrate its superiority across two different outdoor Lidar datasets.

Nermin Samet, Gilles Puy, Oriane Siméoni et al.

Annotating lidar point clouds for autonomous driving is a notoriously expensive and time-consuming task. In this work, we show that the quality of recent self-supervised lidar scan representations allows a great reduction of the annotation cost. Our method has two main steps. First, we show that self-supervised representations allow a simple and direct selection of highly informative lidar scans to annotate: training a network on these selected scans leads to much better results than a random selection of scans and, more interestingly, to results on par with selections made by SOTA active learning methods. In a second step, we leverage the same self-supervised representations to cluster points in our selected scans. Asking the annotator to classify each cluster, with a single click per cluster, then permits us to close the gap with fully-annotated training sets, while only requiring one thousandth of the point labels.

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.

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.

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

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

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

Ellington Kirby, Mickael Chen, Renaud Marlet et al.

The generation of LiDAR scans is a growing topic with diverse applications to autonomous driving. However, scan generation remains challenging, especially when compared to the rapid advancement of image and 3D object generation. We consider the task of LiDAR object generation, requiring models to produce 3D objects as viewed by a LiDAR scan. It focuses LiDAR scan generation on a key aspect of scenes, the objects, while also benefiting from advancements in 3D object generative methods. We introduce a novel diffusion-based model to produce LiDAR point clouds of dataset objects, including intensity, and with an extensive control of the generation via conditioning information. Our experiments on nuScenes and KITTI-360 show the quality of our generations measured with new 3D metrics developed to suit LiDAR objects. The code is available at https://github.com/valeoai/LOGen.

Corentin Sautier, Gilles Puy, Alexandre Boulch et al.

Panoptic segmentation of LiDAR point clouds is fundamental to outdoor scene understanding, with autonomous driving being a primary application. While state-of-the-art approaches typically rely on end-to-end deep learning architectures and extensive manual annotations of instances, the significant cost and time investment required for labeling large-scale point cloud datasets remains a major bottleneck in this field. In this work, we demonstrate that competitive panoptic segmentation can be achieved using only semantic labels, with instances predicted without any training or annotations. Our method outperforms {most} state-of-the-art supervised methods on standard benchmarks including SemanticKITTI and nuScenes, and outperforms every publicly available method on SemanticKITTI as a drop-in instance head replacement, while running in real-time on a single-threaded CPU and requiring no instance labels. It is fully explainable, and requires no learning or parameter tuning. Alpine combined with state-of-the-art semantic segmentation ranks first on the official panoptic segmentation leaderboard of SemanticKITTI. Code is available at https://github.com/valeoai/Alpine/

Björn Michele, Alexandre Boulch, Gilles Puy et al.

Semantic segmentation networks trained under full supervision for one type of lidar fail to generalize to unseen lidars without intervention. To reduce the performance gap under domain shifts, a recent trend is to leverage vision foundation models (VFMs) providing robust features across domains. In this work, we conduct an exhaustive study to identify recipes for exploiting VFMs in unsupervised domain adaptation for semantic segmentation of lidar point clouds. Building upon unsupervised image-to-lidar knowledge distillation, our study reveals that: (1) the architecture of the lidar backbone is key to maximize the generalization performance on a target domain; (2) it is possible to pretrain a single backbone once and for all, and use it to address many domain shifts; (3) best results are obtained by keeping the pretrained backbone frozen and training an MLP head for semantic segmentation. The resulting pipeline achieves state-of-the-art results in four widely-recognized and challenging settings. The code will be available at: https://github.com/valeoai/muddos.

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.

Raphael Sulzer, Loic Landrieu, Alexandre Boulch et al.

Most current neural networks for reconstructing surfaces from point clouds ignore sensor poses and only operate on raw point locations. Sensor visibility, however, holds meaningful information regarding space occupancy and surface orientation. In this paper, we present two simple ways to augment raw point clouds with visibility information, so it can directly be leveraged by surface reconstruction networks with minimal adaptation. Our proposed modifications consistently improve the accuracy of generated surfaces as well as the generalization ability of the networks to unseen shape domains. Our code and data is available at https://github.com/raphaelsulzer/dsrv-data.

Alexandre Boulch, Renaud Marlet

Implicit neural networks have been successfully used for surface reconstruction from point clouds. However, many of them face scalability issues as they encode the isosurface function of a whole object or scene into a single latent vector. To overcome this limitation, a few approaches infer latent vectors on a coarse regular 3D grid or on 3D patches, and interpolate them to answer occupancy queries. In doing so, they loose the direct connection with the input points sampled on the surface of objects, and they attach information uniformly in space rather than where it matters the most, i.e., near the surface. Besides, relying on fixed patch sizes may require discretization tuning. To address these issues, we propose to use point cloud convolutions and compute latent vectors at each input point. We then perform a learning-based interpolation on nearest neighbors using inferred weights. Experiments on both object and scene datasets show that our approach significantly outperforms other methods on most classical metrics, producing finer details and better reconstructing thinner volumes. The code is available at https://github.com/valeoai/POCO.

Anh-Quan Cao, Gilles Puy, Alexandre Boulch et al.

Rigid registration of point clouds with partial overlaps is a longstanding problem usually solved in two steps: (a) finding correspondences between the point clouds; (b) filtering these correspondences to keep only the most reliable ones to estimate the transformation. Recently, several deep nets have been proposed to solve these steps jointly. We built upon these works and propose PCAM: a neural network whose key element is a pointwise product of cross-attention matrices that permits to mix both low-level geometric and high-level contextual information to find point correspondences. These cross-attention matrices also permits the exchange of context information between the point clouds, at each layer, allowing the network construct better matching features within the overlapping regions. The experiments show that PCAM achieves state-of-the-art results among methods which, like us, solve steps (a) and (b) jointly via deepnets. Our code and trained models are available at https://github.com/valeoai/PCAM.

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.

Raphael Sulzer, Loic Landrieu, Renaud Marlet et al.

We introduce a novel learning-based, visibility-aware, surface reconstruction method for large-scale, defect-laden point clouds. Our approach can cope with the scale and variety of point cloud defects encountered in real-life Multi-View Stereo (MVS) acquisitions. Our method relies on a 3D Delaunay tetrahedralization whose cells are classified as inside or outside the surface by a graph neural network and an energy model solvable with a graph cut. Our model, making use of both local geometric attributes and line-of-sight visibility information, is able to learn a visibility model from a small amount of synthetic training data and generalizes to real-life acquisitions. Combining the efficiency of deep learning methods and the scalability of energy based models, our approach outperforms both learning and non learning-based reconstruction algorithms on two publicly available reconstruction benchmarks. Our code and data is available at https://github.com/raphaelsulzer/dgnn.

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.

Tetiana Martyniuk, Gilles Puy, Alexandre Boulch et al.

Training diffusion models that work directly on lidar points at the scale of outdoor scenes is challenging due to the difficulty of generating fine-grained details from white noise over a broad field of view. The latest works addressing scene completion with diffusion models tackle this problem by reformulating the original DDPM as a local diffusion process. It contrasts with the common practice of operating at the level of objects, where vanilla DDPMs are currently used. In this work, we close the gap between these two lines of work. We identify approximations in the local diffusion formulation, show that they are not required to operate at the scene level, and that a vanilla DDPM with a well-chosen starting point is enough for completion. Finally, we demonstrate that our method, LiDPM, leads to better results in scene completion on SemanticKITTI. The project page is https://astra-vision.github.io/LiDPM .

Nermin Samet, Gilles Puy, Renaud Marlet

We study the use of image-based Vision-Language Models (VLMs) for open-vocabulary segmentation of lidar scans in driving settings. Classically, image semantics can be back-projected onto 3D point clouds. Yet, resulting point labels are noisy and sparse. We consolidate these labels to enforce both spatio-temporal consistency and robustness to image-level augmentations. We then train a 3D network based on these refined labels. This simple method, called LOSC, outperforms the SOTA of zero-shot open-vocabulary semantic and panoptic segmentation on both nuScenes and SemanticKITTI, with significant margins.

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.

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.

Alexandre Boulch, Pierre-Alain Langlois, Gilles Puy et al.

There has been recently a growing interest for implicit shape representations. Contrary to explicit representations, they have no resolution limitations and they easily deal with a wide variety of surface topologies. To learn these implicit representations, current approaches rely on a certain level of shape supervision (e.g., inside/outside information or distance-to-shape knowledge), or at least require a dense point cloud (to approximate well enough the distance-to-shape). In contrast, we introduce NeeDrop, a self-supervised method for learning shape representations from possibly extremely sparse point clouds. Like in Buffon's needle problem, we "drop" (sample) needles on the point cloud and consider that, statistically, close to the surface, the needle end points lie on opposite sides of the surface. No shape knowledge is required and the point cloud can be highly sparse, e.g., as lidar point clouds acquired by vehicles. Previous self-supervised shape representation approaches fail to produce good-quality results on this kind of data. We obtain quantitative results on par with existing supervised approaches on shape reconstruction datasets and show promising qualitative results on hard autonomous driving datasets such as KITTI.

Björn Michele, Alexandre Boulch, Gilles Puy et al.

While there has been a number of studies on Zero-Shot Learning (ZSL) for 2D images, its application to 3D data is still recent and scarce, with just a few methods limited to classification. We present the first generative approach for both ZSL and Generalized ZSL (GZSL) on 3D data, that can handle both classification and, for the first time, semantic segmentation. We show that it reaches or outperforms the state of the art on ModelNet40 classification for both inductive ZSL and inductive GZSL. For semantic segmentation, we created three benchmarks for evaluating this new ZSL task, using S3DIS, ScanNet and SemanticKITTI. Our experiments show that our method outperforms strong baselines, which we additionally propose for this task.

Yang Xiao, Yuming Du, Renaud Marlet

Motivated by the need for estimating the 3D pose of arbitrary objects, we consider the challenging problem of class-agnostic object viewpoint estimation from images only, without CAD model knowledge. The idea is to leverage features learned on seen classes to estimate the pose for classes that are unseen, yet that share similar geometries and canonical frames with seen classes. We train a direct pose estimator in a class-agnostic way by sharing weights across all object classes, and we introduce a contrastive learning method that has three main ingredients: (i) the use of pre-trained, self-supervised, contrast-based features; (ii) pose-aware data augmentations; (iii) a pose-aware contrastive loss. We experimented on Pascal3D+, ObjectNet3D and Pix3D in a cross-dataset fashion, with both seen and unseen classes. We report state-of-the-art results, including against methods that additionally use CAD models as input.

Yang Xiao, Vincent Lepetit, Renaud Marlet

Detecting objects and estimating their viewpoints in images are key tasks of 3D scene understanding. Recent approaches have achieved excellent results on very large benchmarks for object detection and viewpoint estimation. However, performances are still lagging behind for novel object categories with few samples. In this paper, we tackle the problems of few-shot object detection and few-shot viewpoint estimation. We demonstrate on both tasks the benefits of guiding the network prediction with class-representative features extracted from data in different modalities: image patches for object detection, and aligned 3D models for viewpoint estimation. Despite its simplicity, our method outperforms state-of-the-art methods by a large margin on a range of datasets, including PASCAL and COCO for few-shot object detection, and Pascal3D+ and ObjectNet3D for few-shot viewpoint estimation. Furthermore, when the 3D model is not available, we introduce a simple category-agnostic viewpoint estimation method by exploiting geometrical similarities and consistent pose labelling across different classes. While it moderately reduces performance, this approach still obtains better results than previous methods in this setting. Last, for the first time, we tackle the combination of both few-shot tasks, on three challenging benchmarks for viewpoint estimation in the wild, ObjectNet3D, Pascal3D+ and Pix3D, showing very promising results.

Xuchong Qiu, Yang Xiao, Chaohui Wang et al.

We formalize concepts around geometric occlusion in 2D images (i.e., ignoring semantics), and propose a novel unified formulation of both occlusion boundaries and occlusion orientations via a pixel-pair occlusion relation. The former provides a way to generate large-scale accurate occlusion datasets while, based on the latter, we propose a novel method for task-independent pixel-level occlusion relationship estimation from single images. Experiments on a variety of datasets demonstrate that our method outperforms existing ones on this task. To further illustrate the value of our formulation, we also propose a new depth map refinement method that consistently improve the performance of state-of-the-art monocular depth estimation methods. Our code and data are available at http://imagine.enpc.fr/~qiux/P2ORM/.

Gilles Puy, Alexandre Boulch, Renaud Marlet

We propose and study a method called FLOT that estimates scene flow on point clouds. We start the design of FLOT by noticing that scene flow estimation on point clouds reduces to estimating a permutation matrix in a perfect world. Inspired by recent works on graph matching, we build a method to find these correspondences by borrowing tools from optimal transport. Then, we relax the transport constraints to take into account real-world imperfections. The transport cost between two points is given by the pairwise similarity between deep features extracted by a neural network trained under full supervision using synthetic datasets. Our main finding is that FLOT can perform as well as the best existing methods on synthetic and real-world datasets while requiring much less parameters and without using multiscale analysis. Our second finding is that, on the training datasets considered, most of the performance can be explained by the learned transport cost. This yields a simpler method, FLOT$_0$, which is obtained using a particular choice of optimal transport parameters and performs nearly as well as FLOT.

Simon Roburin, Yann de Mont-Marin, Andrei Bursuc et al.

Normalization Layers (NLs) are widely used in modern deep-learning architectures. Despite their apparent simplicity, their effect on optimization is not yet fully understood. This paper introduces a spherical framework to study the optimization of neural networks with NLs from a geometric perspective. Concretely, the radial invariance of groups of parameters, such as filters for convolutional neural networks, allows to translate the optimization steps on the $L_2$ unit hypersphere. This formulation and the associated geometric interpretation shed new light on the training dynamics. Firstly, the first effective learning rate expression of Adam is derived. Then the demonstration that, in the presence of NLs, performing Stochastic Gradient Descent (SGD) alone is actually equivalent to a variant of Adam constrained to the unit hypersphere, stems from the framework. Finally, this analysis outlines phenomena that previous variants of Adam act on and their importance in the optimization process are experimentally validated.

Alexandre Boulch, Gilles Puy, Renaud Marlet

Recent state-of-the-art methods for point cloud processing are based on the notion of point convolution, for which several approaches have been proposed. In this paper, inspired by discrete convolution in image processing, we provide a formulation to relate and analyze a number of point convolution methods. We also propose our own convolution variant, that separates the estimation of geometry-less kernel weights and their alignment to the spatial support of features. Additionally, we define a point sampling strategy for convolution that is both effective and fast. Finally, using our convolution and sampling strategy, we show competitive results on classification and semantic segmentation benchmarks while being time and memory efficient.

Pierre-Alain Langlois, Alexandre Boulch, Renaud Marlet

In man-made environments such as indoor scenes, when point-based 3D reconstruction fails due to the lack of texture, lines can still be detected and used to support surfaces. We present a novel method for watertight piecewise-planar surface reconstruction from 3D line segments with visibility information. First, planes are extracted by a novel RANSAC approach for line segments that allows multiple shape support. Then, each 3D cell of a plane arrangement is labeled full or empty based on line attachment to planes, visibility and regularization. Experiments show the robustness to sparse input data, noise and outliers.

Yang Xiao, Xuchong Qiu, Pierre-Alain Langlois et al.

Most deep pose estimation methods need to be trained for specific object instances or categories. In this work we propose a completely generic deep pose estimation approach, which does not require the network to have been trained on relevant categories, nor objects in a category to have a canonical pose. We believe this is a crucial step to design robotic systems that can interact with new objects in the wild not belonging to a predefined category. Our main insight is to dynamically condition pose estimation with a representation of the 3D shape of the target object. More precisely, we train a Convolutional Neural Network that takes as input both a test image and a 3D model, and outputs the relative 3D pose of the object in the input image with respect to the 3D model. We demonstrate that our method boosts performances for supervised category pose estimation on standard benchmarks, namely Pascal3D+, ObjectNet3D and Pix3D, on which we provide results superior to the state of the art. More importantly, we show that our network trained on everyday man-made objects from ShapeNet generalizes without any additional training to completely new types of 3D objects by providing results on the LINEMOD dataset as well as on natural entities such as animals from ImageNet.

Vianney Loing, Renaud Marlet, Mathieu Aubry

Localizing an object accurately with respect to a robot is a key step for autonomous robotic manipulation. In this work, we propose to tackle this task knowing only 3D models of the robot and object in the particular case where the scene is viewed from uncalibrated cameras -- a situation which would be typical in an uncontrolled environment, e.g., on a construction site. We demonstrate that this localization can be performed very accurately, with millimetric errors, without using a single real image for training, a strong advantage since acquiring representative training data is a long and expensive process. Our approach relies on a classification Convolutional Neural Network (CNN) trained using hundreds of thousands of synthetically rendered scenes with randomized parameters. To evaluate our approach quantitatively and make it comparable to alternative approaches, we build a new rich dataset of real robot images with accurately localized blocks.

Yohann Salaun, Renaud Marlet, Pascal Monasse

Usual Structure-from-Motion (SfM) techniques require at least trifocal overlaps to calibrate cameras and reconstruct a scene. We consider here scenarios of reduced image sets with little overlap, possibly as low as two images at most seeing the same part of the scene. We propose a new method, based on line coplanarity hypotheses, for estimating the relative scale of two independent bifocal calibrations sharing a camera, without the need of any trifocal information or Manhattan-world assumption. We use it to compute SfM in a chain of up-to-scale relative motions. For accuracy, we however also make use of trifocal information for line and/or point features, when present, relaxing usual trifocal constraints. For robustness to wrong assumptions and mismatches, we embed all constraints in a parameterless RANSAC-like approach. Experiments show that we can calibrate datasets that previously could not, and that this wider applicability does not come at the cost of inaccuracy.

Francisco Massa, Renaud Marlet, Mathieu Aubry

Convolutional Neural Networks (CNNs) were recently shown to provide state-of-the-art results for object category viewpoint estimation. However different ways of formulating this problem have been proposed and the competing approaches have been explored with very different design choices. This paper presents a comparison of these approaches in a unified setting as well as a detailed analysis of the key factors that impact performance. Followingly, we present a new joint training method with the detection task and demonstrate its benefit. We also highlight the superiority of classification approaches over regression approaches, quantify the benefits of deeper architectures and extended training data, and demonstrate that synthetic data is beneficial even when using ImageNet training data. By combining all these elements, we demonstrate an improvement of approximately 5% mAVP over previous state-of-the-art results on the Pascal3D+ dataset. In particular for their most challenging 24 view classification task we improve the results from 31.1% to 36.1% mAVP.

Raghudeep Gadde, Varun Jampani, Renaud Marlet et al.

This paper introduces a fast and efficient segmentation technique for 2D images and 3D point clouds of building facades. Facades of buildings are highly structured and consequently most methods that have been proposed for this problem aim to make use of this strong prior information. Contrary to most prior work, we are describing a system that is almost domain independent and consists of standard segmentation methods. We train a sequence of boosted decision trees using auto-context features. This is learned using stacked generalization. We find that this technique performs better, or comparable with all previous published methods and present empirical results on all available 2D and 3D facade benchmark datasets. The proposed method is simple to implement, easy to extend, and very efficient at test-time inference.

Francisco Massa, Mathieu Aubry, Renaud Marlet

In this paper we study the application of convolutional neural networks for jointly detecting objects depicted in still images and estimating their 3D pose. We identify different feature representations of oriented objects, and energies that lead a network to learn this representations. The choice of the representation is crucial since the pose of an object has a natural, continuous structure while its category is a discrete variable. We evaluate the different approaches on the joint object detection and pose estimation task of the Pascal3D+ benchmark using Average Viewpoint Precision. We show that a classification approach on discretized viewpoints achieves state-of-the-art performance for joint object detection and pose estimation, and significantly outperforms existing baselines on this benchmark.