Mathieu Aubry

Tom Monnier, Matthew Fisher, Alexei A. Efros et al. · berkeley

Approaches for single-view reconstruction typically rely on viewpoint annotations, silhouettes, the absence of background, multiple views of the same instance, a template shape, or symmetry. We avoid all such supervision and assumptions by explicitly leveraging the consistency between images of different object instances. As a result, our method can learn from large collections of unlabelled images depicting the same object category. Our main contributions are two ways for leveraging cross-instance consistency: (i) progressive conditioning, a training strategy to gradually specialize the model from category to instances in a curriculum learning fashion; and (ii) neighbor reconstruction, a loss enforcing consistency between instances having similar shape or texture. Also critical to the success of our method are: our structured autoencoding architecture decomposing an image into explicit shape, texture, pose, and background; an adapted formulation of differential rendering; and a new optimization scheme alternating between 3D and pose learning. We compare our approach, UNICORN, both on the diverse synthetic ShapeNet dataset - the classical benchmark for methods requiring multiple views as supervision - and on standard real-image benchmarks (Pascal3D+ Car, CUB) for which most methods require known templates and silhouette annotations. We also showcase applicability to more challenging real-world collections (CompCars, LSUN), where silhouettes are not available and images are not cropped around the object.

Tom Monnier, Jake Austin, Angjoo Kanazawa et al. · berkeley

Given a set of calibrated images of a scene, we present an approach that produces a simple, compact, and actionable 3D world representation by means of 3D primitives. While many approaches focus on recovering high-fidelity 3D scenes, we focus on parsing a scene into mid-level 3D representations made of a small set of textured primitives. Such representations are interpretable, easy to manipulate and suited for physics-based simulations. Moreover, unlike existing primitive decomposition methods that rely on 3D input data, our approach operates directly on images through differentiable rendering. Specifically, we model primitives as textured superquadric meshes and optimize their parameters from scratch with an image rendering loss. We highlight the importance of modeling transparency for each primitive, which is critical for optimization and also enables handling varying numbers of primitives. We show that the resulting textured primitives faithfully reconstruct the input images and accurately model the visible 3D points, while providing amodal shape completions of unseen object regions. We compare our approach to the state of the art on diverse scenes from DTU, and demonstrate its robustness on real-life captures from BlendedMVS and Nerfstudio. We also showcase how our results can be used to effortlessly edit a scene or perform physical simulations. Code and video results are available at https://www.tmonnier.com/DBW .

Ioannis Siglidis, Aleksander Holynski, Alexei A. Efros et al. · berkeley

This paper demonstrates how to use generative models trained for image synthesis as tools for visual data mining. Our insight is that since contemporary generative models learn an accurate representation of their training data, we can use them to summarize the data by mining for visual patterns. Concretely, we show that after finetuning conditional diffusion models to synthesize images from a specific dataset, we can use these models to define a typicality measure on that dataset. This measure assesses how typical visual elements are for different data labels, such as geographic location, time stamps, semantic labels, or even the presence of a disease. This analysis-by-synthesis approach to data mining has two key advantages. First, it scales much better than traditional correspondence-based approaches since it does not require explicitly comparing all pairs of visual elements. Second, while most previous works on visual data mining focus on a single dataset, our approach works on diverse datasets in terms of content and scale, including a historical car dataset, a historical face dataset, a large worldwide street-view dataset, and an even larger scene dataset. Furthermore, our approach allows for translating visual elements across class labels and analyzing consistent changes.

Yann Labbé, Lucas Manuelli, Arsalan Mousavian et al.

We introduce MegaPose, a method to estimate the 6D pose of novel objects, that is, objects unseen during training. At inference time, the method only assumes knowledge of (i) a region of interest displaying the object in the image and (ii) a CAD model of the observed object. The contributions of this work are threefold. First, we present a 6D pose refiner based on a render&compare strategy which can be applied to novel objects. The shape and coordinate system of the novel object are provided as inputs to the network by rendering multiple synthetic views of the object's CAD model. Second, we introduce a novel approach for coarse pose estimation which leverages a network trained to classify whether the pose error between a synthetic rendering and an observed image of the same object can be corrected by the refiner. Third, we introduce a large-scale synthetic dataset of photorealistic images of thousands of objects with diverse visual and shape properties and show that this diversity is crucial to obtain good generalization performance on novel objects. We train our approach on this large synthetic dataset and apply it without retraining to hundreds of novel objects in real images from several pose estimation benchmarks. Our approach achieves state-of-the-art performance on the ModelNet and YCB-Video datasets. An extensive evaluation on the 7 core datasets of the BOP challenge demonstrates that our approach achieves performance competitive with existing approaches that require access to the target objects during training. Code, dataset and trained models are available on the project page: https://megapose6d.github.io/.

Raphael Baena, Syrine Kalleli, Mathieu Aubry

We introduce a general detection-based approach to text line recognition, be it printed (OCR) or handwritten (HTR), with Latin, Chinese, or ciphered characters. Detection-based approaches have until now been largely discarded for HTR because reading characters separately is often challenging, and character-level annotation is difficult and expensive. We overcome these challenges thanks to three main insights: (i) synthetic pre-training with sufficiently diverse data enables learning reasonable character localization for any script; (ii) modern transformer-based detectors can jointly detect a large number of instances, and, if trained with an adequate masking strategy, leverage consistency between the different detections; (iii) once a pre-trained detection model with approximate character localization is available, it is possible to fine-tune it with line-level annotation on real data, even with a different alphabet. Our approach, dubbed DTLR, builds on a completely different paradigm than state-of-the-art HTR methods, which rely on autoregressive decoding, predicting character values one by one, while we treat a complete line in parallel. Remarkably, we demonstrate good performance on a large range of scripts, usually tackled with specialized approaches. In particular, we improve state-of-the-art performances for Chinese script recognition on the CASIA v2 dataset, and for cipher recognition on the Borg and Copiale datasets. Our code and models are available at https://github.com/raphael-baena/DTLR.

Georgy Ponimatkin, Yann Labbé, Bryan Russell et al.

We introduce FocalPose, a neural render-and-compare method for jointly estimating the camera-object 6D pose and camera focal length given a single RGB input image depicting a known object. The contributions of this work are twofold. First, we derive a focal length update rule that extends an existing state-of-the-art render-and-compare 6D pose estimator to address the joint estimation task. Second, we investigate several different loss functions for jointly estimating the object pose and focal length. We find that a combination of direct focal length regression with a reprojection loss disentangling the contribution of translation, rotation, and focal length leads to improved results. We show results on three challenging benchmark datasets that depict known 3D models in uncontrolled settings. We demonstrate that our focal length and 6D pose estimates have lower error than the existing state-of-the-art methods.

Elliot Vincent, Jean Ponce, Mathieu Aubry

Improvements in Earth observation by satellites allow for imagery of ever higher temporal and spatial resolution. Leveraging this data for agricultural monitoring is key for addressing environmental and economic challenges. Current methods for crop segmentation using temporal data either rely on annotated data or are heavily engineered to compensate the lack of supervision. In this paper, we present and compare datasets and methods for both supervised and unsupervised pixel-wise segmentation of satellite image time series (SITS). We also introduce an approach to add invariance to spectral deformations and temporal shifts to classical prototype-based methods such as K-means and Nearest Centroid Classifier (NCC). We study different levels of supervision and show this simple and highly interpretable method achieves the best performance in the low data regime and significantly improves the state of the art for unsupervised classification of agricultural time series on four recent SITS datasets.

Romain Loiseau, Mathieu Aubry, Loïc Landrieu

Roof-mounted spinning LiDAR sensors are widely used by autonomous vehicles. However, most semantic datasets and algorithms used for LiDAR sequence segmentation operate on $360^\circ$ frames, causing an acquisition latency incompatible with real-time applications. To address this issue, we first introduce HelixNet, a $10$ billion point dataset with fine-grained labels, timestamps, and sensor rotation information necessary to accurately assess the real-time readiness of segmentation algorithms. Second, we propose Helix4D, a compact and efficient spatio-temporal transformer architecture specifically designed for rotating LiDAR sequences. Helix4D operates on acquisition slices corresponding to a fraction of a full sensor rotation, significantly reducing the total latency. Helix4D reaches accuracy on par with the best segmentation algorithms on HelixNet and SemanticKITTI with a reduction of over $5\times$ in terms of latency and $50\times$ in model size. The code and data are available at: https://romainloiseau.fr/helixnet

Elliot Vincent, Jean Ponce, Mathieu Aubry

Satellite imagery plays a crucial role in monitoring changes happening on Earth's surface and aiding in climate analysis, ecosystem assessment, and disaster response. In this paper, we tackle semantic change detection with satellite image time series (SITS-SCD) which encompasses both change detection and semantic segmentation tasks. We propose a new architecture that improves over the state of the art, scales better with the number of parameters, and leverages long-term temporal information. However, for practical use cases, models need to adapt to spatial and temporal shifts, which remains a challenge. We investigate the impact of temporal and spatial shifts separately on global, multi-year SITS datasets using DynamicEarthNet and MUDS. We show that the spatial domain shift represents the most complex setting and that the impact of temporal shift on performance is more pronounced on change detection than on semantic segmentation, highlighting that it is a specific issue deserving further attention.

Romain Loiseau, Elliot Vincent, Mathieu Aubry et al.

We propose an unsupervised method for parsing large 3D scans of real-world scenes with easily-interpretable shapes. This work aims to provide a practical tool for analyzing 3D scenes in the context of aerial surveying and mapping, without the need for user annotations. Our approach is based on a probabilistic reconstruction model that decomposes an input 3D point cloud into a small set of learned prototypical 3D shapes. The resulting reconstruction is visually interpretable and can be used to perform unsupervised instance and low-shot semantic segmentation of complex scenes. We demonstrate the usefulness of our model on a novel dataset of seven large aerial LiDAR scans from diverse real-world scenarios. Our approach outperforms state-of-the-art unsupervised methods in terms of decomposition accuracy while remaining visually interpretable. Our code and dataset are available at https://romainloiseau.fr/learnable-earth-parser/

Martin Cífka, Georgy Ponimatkin, Yann Labbé et al.

We introduce FocalPose++, a neural render-and-compare method for jointly estimating the camera-object 6D pose and camera focal length given a single RGB input image depicting a known object. The contributions of this work are threefold. First, we derive a focal length update rule that extends an existing state-of-the-art render-and-compare 6D pose estimator to address the joint estimation task. Second, we investigate several different loss functions for jointly estimating the object pose and focal length. We find that a combination of direct focal length regression with a reprojection loss disentangling the contribution of translation, rotation, and focal length leads to improved results. Third, we explore the effect of different synthetic training data on the performance of our method. Specifically, we investigate different distributions used for sampling object's 6D pose and camera's focal length when rendering the synthetic images, and show that parametric distribution fitted on real training data works the best. We show results on three challenging benchmark datasets that depict known 3D models in uncontrolled settings. We demonstrate that our focal length and 6D pose estimates have lower error than the existing state-of-the-art methods.

Theo Deprelle, Thibault Groueix, Noam Aigerman et al.

This paper describes new techniques for learning atlas-like representations of 3D surfaces, i.e. homeomorphic transformations from a 2D domain to surfaces. Compared to prior work, we propose two major contributions. First, instead of mapping a fixed 2D domain, such as a set of square patches, to the surface, we learn a continuous 2D domain with arbitrary topology by optimizing a point sampling distribution represented as a mixture of Gaussians. Second, we learn consistent mappings in both directions: charts, from the 3D surface to 2D domain, and parametrizations, their inverse. We demonstrate that this improves the quality of the learned surface representation, as well as its consistency in a collection of related shapes. It thus leads to improvements for applications such as correspondence estimation, texture transfer, and consistent UV mapping. As an additional technical contribution, we outline that, while incorporating normal consistency has clear benefits, it leads to issues in the optimization, and that these issues can be mitigated using a simple repulsive regularization. We demonstrate that our contributions provide better surface representation than existing baselines.

Elliot Vincent, Mehraïl Saroufim, Jonathan Chemla et al.

Archaeological sites are the physical remains of past human activity and one of the main sources of information about past societies and cultures. However, they are also the target of malevolent human actions, especially in countries having experienced inner turmoil and conflicts. Because monitoring these sites from space is a key step towards their preservation, we introduce the DAFA Looted Sites dataset, \datasetname, a labeled multi-temporal remote sensing dataset containing 55,480 images acquired monthly over 8 years across 675 Afghan archaeological sites, including 135 sites looted during the acquisition period. \datasetname~is particularly challenging because of the limited number of training samples, the class imbalance, the weak binary annotations only available at the level of the time series, and the subtlety of relevant changes coupled with important irrelevant ones over a long time period. It is also an interesting playground to assess the performance of satellite image time series (SITS) classification methods on a real and important use case. We evaluate a large set of baselines, outline the substantial benefits of using foundation models and show the additional boost that can be provided by using complete time series instead of using a single image.

Ioannis Siglidis, Nicolas Gonthier, Julien Gaubil et al.

We present a generative document-specific approach to character analysis and recognition in text lines. Our main idea is to build on unsupervised multi-object segmentation methods and in particular those that reconstruct images based on a limited amount of visual elements, called sprites. Taking as input a set of text lines with similar font or handwriting, our approach can learn a large number of different characters and leverage line-level annotations when available. Our contribution is twofold. First, we provide the first adaptation and evaluation of a deep unsupervised multi-object segmentation approach for text line analysis. Since these methods have mainly been evaluated on synthetic data in a completely unsupervised setting, demonstrating that they can be adapted and quantitatively evaluated on real images of text and that they can be trained using weak supervision are significant progresses. Second, we show the potential of our method for new applications, more specifically in the field of paleography, which studies the history and variations of handwriting, and for cipher analysis. We demonstrate our approach on three very different datasets: a printed volume of the Google1000 dataset, the Copiale cipher and historical handwritten charters from the 12th and early 13th century.

Sayan Kumar Chaki, Zeynep Sonat Baltaci, Elliot Vincent et al.

This paper aims to develop the study of historical printed ornaments with modern unsupervised computer vision. We highlight three complex tasks that are of critical interest to book historians: clustering, element discovery, and unsupervised change localization. For each of these tasks, we introduce an evaluation benchmark, and we adapt and evaluate state-of-the-art models. Our Rey's Ornaments dataset is designed to be a representative example of a set of ornaments historians would be interested in. It focuses on an XVIIIth century bookseller, Marc-Michel Rey, providing a consistent set of ornaments with a wide diversity and representative challenges. Our results highlight the limitations of state-of-the-art models when faced with real data and show simple baselines such as k-means or congealing can outperform more sophisticated approaches on such data. Our dataset and code can be found at https://printed-ornaments.github.io/.

Marta López-Rauhut, Loic Landrieu, Mathieu Aubry et al.

New multi-modal large language models (MLLMs) are continuously being trained and deployed, following rapid development cycles. This generative AI frenzy is driving steady increases in energy consumption, greenhouse gas emissions, and a plethora of other environmental impacts linked to datacenter construction and hardware manufacturing. Mitigating the environmental consequences of GenAI remains challenging due to an overall lack of transparency by the main actors in the field. Even when the environmental impacts of specific models are mentioned, they are typically restricted to the carbon footprint of the final training run, omitting the research and development stages. In this work, we explore the impact of GenAI research through a fine-grained analysis of the compute spent to create Moshi, a 7B-parameter speech-text foundation model for real-time dialogue developed by Kyutai, a leading privately funded open science AI lab. For the first time, our study dives into the anatomy of compute-intensive MLLM research, quantifying the GPU-time invested in specific model components and training phases, as well as early experimental stages, failed training runs, debugging, and ablation studies. Additionally, we assess the environmental impacts of creating Moshi from beginning to end using a life cycle assessment methodology: we quantify energy and water consumption, greenhouse gas emissions, and mineral resource depletion associated with the production and use of datacenter hardware. Our detailed analysis allows us to provide actionable guidelines to reduce compute usage and environmental impacts of MLLM research, paving the way for more sustainable AI research.

Marta López-Rauhut, Hongyu Zhou, Mathieu Aubry et al.

Historical maps offer an invaluable perspective into territory evolution across past centuries--long before satellite or remote sensing technologies existed. Deep learning methods have shown promising results in segmenting historical maps, but publicly available datasets typically focus on a single map type or period, require extensive and costly annotations, and are not suited for nationwide, long-term analyses. In this paper, we introduce a new dataset of historical maps tailored for analyzing large-scale, long-term land use and land cover evolution with limited annotations. Spanning metropolitan France (548,305 km^2), our dataset contains three map collections from the 18th, 19th, and 20th centuries. We provide both comprehensive modern labels and 22,878 km^2 of manually annotated historical labels for the 18th and 19th century maps. Our dataset illustrates the complexity of the segmentation task, featuring stylistic inconsistencies, interpretive ambiguities, and significant landscape changes (e.g., marshlands disappearing in favor of forests). We assess the difficulty of these challenges by benchmarking three approaches: a fully-supervised model trained with historical labels, and two weakly-supervised models that rely only on modern annotations. The latter either use the modern labels directly or first perform image-to-image translation to address the stylistic gap between historical and contemporary maps. Finally, we discuss how these methods can support long-term environment monitoring, offering insights into centuries of landscape transformation. Our official project repository is publicly available at https://github.com/Archiel19/FRAx4.git.

Abhishek Kuriyal, Elliot Vincent, Mathieu Aubry et al.

Global variations in terrain appearance raise a major challenge for satellite image analysis, leading to poor model performance when training on locations that differ from those encountered at test time. This remains true even with recent large global datasets. To address this challenge, we propose a novel domain-generalization framework for satellite images. Instead of trying to learn a single generalizable model, we train one expert model per training domain, while learning experts' similarity and encouraging similar experts to be consistent. A model selection module then identifies the most suitable experts for a given test sample and aggregates their predictions. Experiments on four datasets (DynamicEarthNet, MUDS, OSCD, and FMoW) demonstrate consistent gains over existing domain generalization and adaptation methods. Our code is publicly available at https://github.com/Abhishek19009/CoDEx.

Malamatenia Vlachou-Efstathiou, Ioannis Siglidis, Dominique Stutzmann et al.

Defining script types and establishing classification criteria for medieval handwriting is a central aspect of palaeographical analysis. However, existing typologies often encounter methodological challenges, such as descriptive limitations and subjective criteria. We propose an interpretable deep learning-based approach to morphological script type analysis, which enables systematic and objective analysis and contributes to bridging the gap between qualitative observations and quantitative measurements. More precisely, we adapt a deep instance segmentation method to learn comparable character prototypes, representative of letter morphology, and provide qualitative and quantitative tools for their comparison and analysis. We demonstrate our approach by applying it to the Textualis Formata script type and its two subtypes formalized by A. Derolez: Northern and Southern Textualis

Zeynep Sonat Baltacı, Romain Loiseau, Mathieu Aubry

While foundation models drive steady progress in image segmentation and diffusion algorithms compose always more realistic images, the seemingly simple problem of identifying recurrent patterns in a collection of images remains very much open. In this paper, we focus on sprite-based image decomposition models, which have shown some promise for clustering and image decomposition and are appealing because of their high interpretability. These models come in different flavors, need to be tailored to specific datasets, and struggle to scale to images with many objects. We dive into the details of their design, identify their core components, and perform an extensive analysis on clustering benchmarks. We leverage this analysis to propose a deep sprite-based image decomposition method that performs on par with state-of-the-art unsupervised class-aware image segmentation methods on the standard CLEVR benchmark, scales linearly with the number of objects, identifies explicitly object categories, and fully models images in an easily interpretable way.

Syrine Kalleli, Scott Trigg, Ségolène Albouy et al.

Automatically extracting the geometric content from the hundreds of thousands of diagrams drawn in historical manuscripts would enable historians to study the diffusion of astronomical knowledge on a global scale. However, state-of-the-art vectorization methods, often designed to tackle modern data, are not adapted to the complexity and diversity of historical astronomical diagrams. Our contribution is thus twofold. First, we introduce a unique dataset of 303 astronomical diagrams from diverse traditions, ranging from the XIIth to the XVIIIth century, annotated with more than 3000 line segments, circles and arcs. Second, we develop a model that builds on DINO-DETR to enable the prediction of multiple geometric primitives. We show that it can be trained solely on synthetic data and accurately predict primitives on our challenging dataset. Our approach widely improves over the LETR baseline, which is restricted to lines, by introducing a meaningful parametrization for multiple primitives, jointly training for detection and parameter refinement, using deformable attention and training on rich synthetic data. Our dataset and code are available on our webpage.

François Darmon, Bénédicte Bascle, Jean-Clément Devaux et al.

Neural implicit surfaces have become an important technique for multi-view 3D reconstruction but their accuracy remains limited. In this paper, we argue that this comes from the difficulty to learn and render high frequency textures with neural networks. We thus propose to add to the standard neural rendering optimization a direct photo-consistency term across the different views. Intuitively, we optimize the implicit geometry so that it warps views on each other in a consistent way. We demonstrate that two elements are key to the success of such an approach: (i) warping entire patches, using the predicted occupancy and normals of the 3D points along each ray, and measuring their similarity with a robust structural similarity (SSIM); (ii) handling visibility and occlusion in such a way that incorrect warps are not given too much importance while encouraging a reconstruction as complete as possible. We evaluate our approach, dubbed NeuralWarp, on the standard DTU and EPFL benchmarks and show it outperforms state of the art unsupervised implicit surfaces reconstructions by over 20% on both datasets.

Xi Shen, Alexei A. Efros, Armand Joulin et al.

The goal of this work is to efficiently identify visually similar patterns in images, e.g. identifying an artwork detail copied between an engraving and an oil painting, or recognizing parts of a night-time photograph visible in its daytime counterpart. Lack of training data is a key challenge for this co-segmentation task. We present a simple yet surprisingly effective approach to overcome this difficulty: we generate synthetic training pairs by selecting segments in an image and copy-pasting them into another image. We then learn to predict the repeated region masks. We find that it is crucial to predict the correspondences as an auxiliary task and to use Poisson blending and style transfer on the training pairs to generalize on real data. We analyse results with two deep architectures relevant to our joint image analysis task: a transformer-based architecture and Sparse Nc-Net, a recent network designed to predict coarse correspondences using 4D convolutions. We show our approach provides clear improvements for artwork details retrieval on the Brueghel dataset and achieves competitive performance on two place recognition benchmarks, Tokyo247 and Pitts30K. We also demonstrate the potential of our approach for unsupervised image collection analysis by introducing a spectral graph clustering approach to object discovery and demonstrating it on the object discovery dataset of \cite{rubinstein2013unsupervised} and the Brueghel dataset. Our code and data are available at http://imagine.enpc.fr/~shenx/SegSwap/.

Romain Loiseau, Tom Monnier, Mathieu Aubry et al.

In this paper, we revisit the classical representation of 3D point clouds as linear shape models. Our key insight is to leverage deep learning to represent a collection of shapes as affine transformations of low-dimensional linear shape models. Each linear model is characterized by a shape prototype, a low-dimensional shape basis and two neural networks. The networks take as input a point cloud and predict the coordinates of a shape in the linear basis and the affine transformation which best approximate the input. Both linear models and neural networks are learned end-to-end using a single reconstruction loss. The main advantage of our approach is that, in contrast to many recent deep approaches which learn feature-based complex shape representations, our model is explicit and every operation occurs in 3D space. As a result, our linear shape models can be easily visualized and annotated, and failure cases can be visually understood. While our main goal is to introduce a compact and interpretable representation of shape collections, we show it leads to state of the art results for few-shot segmentation.

Ryad Kaoua, Xi Shen, Alexandra Durr et al.

Illustrations are an essential transmission instrument. For an historian, the first step in studying their evolution in a corpus of similar manuscripts is to identify which ones correspond to each other. This image collation task is daunting for manuscripts separated by many lost copies, spreading over centuries, which might have been completely re-organized and greatly modified to adapt to novel knowledge or belief and include hundreds of illustrations. Our contributions in this paper are threefold. First, we introduce the task of illustration collation and a large annotated public dataset to evaluate solutions, including 6 manuscripts of 2 different texts with more than 2 000 illustrations and 1 200 annotated correspondences. Second, we analyze state of the art similarity measures for this task and show that they succeed in simple cases but struggle for large manuscripts when the illustrations have undergone very significant changes and are discriminated only by fine details. Finally, we show clear evidence that significant performance boosts can be expected by exploiting cycle-consistent correspondences. Our code and data are available on http://imagine.enpc.fr/~shenx/ImageCollation.

François Darmon, Bénédicte Bascle, Jean-Clément Devaux et al.

Deep multi-view stereo (MVS) methods have been developed and extensively compared on simple datasets, where they now outperform classical approaches. In this paper, we ask whether the conclusions reached in controlled scenarios are still valid when working with Internet photo collections. We propose a methodology for evaluation and explore the influence of three aspects of deep MVS methods: network architecture, training data, and supervision. We make several key observations, which we extensively validate quantitatively and qualitatively, both for depth prediction and complete 3D reconstructions. First, complex unsupervised approaches cannot train on data in the wild. Our new approach makes it possible with three key elements: upsampling the output, softmin based aggregation and a single reconstruction loss. Second, supervised deep depthmap-based MVS methods are state-of-the art for reconstruction of few internet images. Finally, our evaluation provides very different results than usual ones. This shows that evaluation in uncontrolled scenarios is important for new architectures.

Tom Monnier, Elliot Vincent, Jean Ponce et al.

We present an unsupervised learning framework for decomposing images into layers of automatically discovered object models. Contrary to recent approaches that model image layers with autoencoder networks, we represent them as explicit transformations of a small set of prototypical images. Our model has three main components: (i) a set of object prototypes in the form of learnable images with a transparency channel, which we refer to as sprites; (ii) differentiable parametric functions predicting occlusions and transformation parameters necessary to instantiate the sprites in a given image; (iii) a layered image formation model with occlusion for compositing these instances into complete images including background. By jointly learning the sprites and occlusion/transformation predictors to reconstruct images, our approach not only yields accurate layered image decompositions, but also identifies object categories and instance parameters. We first validate our approach by providing results on par with the state of the art on standard multi-object synthetic benchmarks (Tetrominoes, Multi-dSprites, CLEVR6). We then demonstrate the applicability of our model to real images in tasks that include clustering (SVHN, GTSRB), cosegmentation (Weizmann Horse) and object discovery from unfiltered social network images. To the best of our knowledge, our approach is the first layered image decomposition algorithm that learns an explicit and shared concept of object type, and is robust enough to be applied to real images.

Yann Labbé, Justin Carpentier, Mathieu Aubry et al.

We introduce RoboPose, a method to estimate the joint angles and the 6D camera-to-robot pose of a known articulated robot from a single RGB image. This is an important problem to grant mobile and itinerant autonomous systems the ability to interact with other robots using only visual information in non-instrumented environments, especially in the context of collaborative robotics. It is also challenging because robots have many degrees of freedom and an infinite space of possible configurations that often result in self-occlusions and depth ambiguities when imaged by a single camera. The contributions of this work are three-fold. First, we introduce a new render & compare approach for estimating the 6D pose and joint angles of an articulated robot that can be trained from synthetic data, generalizes to new unseen robot configurations at test time, and can be applied to a variety of robots. Second, we experimentally demonstrate the importance of the robot parametrization for the iterative pose updates and design a parametrization strategy that is independent of the robot structure. Finally, we show experimental results on existing benchmark datasets for four different robots and demonstrate that our method significantly outperforms the state of the art. Code and pre-trained models are available on the project webpage https://www.di.ens.fr/willow/research/robopose/.

Tom Monnier, Mathieu Aubry

We present docExtractor, a generic approach for extracting visual elements such as text lines or illustrations from historical documents without requiring any real data annotation. We demonstrate it provides high-quality performances as an off-the-shelf system across a wide variety of datasets and leads to results on par with state-of-the-art when fine-tuned. We argue that the performance obtained without fine-tuning on a specific dataset is critical for applications, in particular in digital humanities, and that the line-level page segmentation we address is the most relevant for a general purpose element extraction engine. We rely on a fast generator of rich synthetic documents and design a fully convolutional network, which we show to generalize better than a detection-based approach. Furthermore, we introduce a new public dataset dubbed IlluHisDoc dedicated to the fine evaluation of illustration segmentation in historical documents.

François Darmon, Mathieu Aubry, Pascal Monasse

We tackle the problem of finding accurate and robust keypoint correspondences between images. We propose a learning-based approach to guide local feature matches via a learned approximate image matching. Our approach can boost the results of SIFT to a level similar to state-of-the-art deep descriptors, such as Superpoint, ContextDesc, or D2-Net and can improve performance for these descriptors. We introduce and study different levels of supervision to learn coarse correspondences. In particular, we show that weak supervision from epipolar geometry leads to performances higher than the stronger but more biased point level supervision and is a clear improvement over weak image level supervision. We demonstrate the benefits of our approach in a variety of conditions by evaluating our guided keypoint correspondences for localization of internet images on the YFCC100M dataset and indoor images on theSUN3D dataset, for robust localization on the Aachen day-night benchmark and for 3D reconstruction in challenging conditions using the LTLL historical image data.

Yann Labbé, Justin Carpentier, Mathieu Aubry et al.

We introduce an approach for recovering the 6D pose of multiple known objects in a scene captured by a set of input images with unknown camera viewpoints. First, we present a single-view single-object 6D pose estimation method, which we use to generate 6D object pose hypotheses. Second, we develop a robust method for matching individual 6D object pose hypotheses across different input images in order to jointly estimate camera viewpoints and 6D poses of all objects in a single consistent scene. Our approach explicitly handles object symmetries, does not require depth measurements, is robust to missing or incorrect object hypotheses, and automatically recovers the number of objects in the scene. Third, we develop a method for global scene refinement given multiple object hypotheses and their correspondences across views. This is achieved by solving an object-level bundle adjustment problem that refines the poses of cameras and objects to minimize the reprojection error in all views. We demonstrate that the proposed method, dubbed CosyPose, outperforms current state-of-the-art results for single-view and multi-view 6D object pose estimation by a large margin on two challenging benchmarks: the YCB-Video and T-LESS datasets. Code and pre-trained models are available on the project webpage https://www.di.ens.fr/willow/research/cosypose/.

Othman Sbai, Camille Couprie, Mathieu Aubry

The quality and generality of deep image features is crucially determined by the data they have been trained on, but little is known about this often overlooked effect. In this paper, we systematically study the effect of variations in the training data by evaluating deep features trained on different image sets in a few-shot classification setting. The experimental protocol we define allows to explore key practical questions. What is the influence of the similarity between base and test classes? Given a fixed annotation budget, what is the optimal trade-off between the number of images per class and the number of classes? Given a fixed dataset, can features be improved by splitting or combining different classes? Should simple or diverse classes be annotated? In a wide range of experiments, we provide clear answers to these questions on the miniImageNet, ImageNet and CUB-200 benchmarks. We also show how the base dataset design can improve performance in few-shot classification more drastically than replacing a simple baseline by an advanced state of the art algorithm.

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.

Tom Monnier, Thibault Groueix, Mathieu Aubry

Recent advances in image clustering typically focus on learning better deep representations. In contrast, we present an orthogonal approach that does not rely on abstract features but instead learns to predict image transformations and performs clustering directly in image space. This learning process naturally fits in the gradient-based training of K-means and Gaussian mixture model, without requiring any additional loss or hyper-parameters. It leads us to two new deep transformation-invariant clustering frameworks, which jointly learn prototypes and transformations. More specifically, we use deep learning modules that enable us to resolve invariance to spatial, color and morphological transformations. Our approach is conceptually simple and comes with several advantages, including the possibility to easily adapt the desired invariance to the task and a strong interpretability of both cluster centers and assignments to clusters. We demonstrate that our novel approach yields competitive and highly promising results on standard image clustering benchmarks. Finally, we showcase its robustness and the advantages of its improved interpretability by visualizing clustering results over real photograph collections.

Xi Shen, François Darmon, Alexei A. Efros et al.

This paper considers the generic problem of dense alignment between two images, whether they be two frames of a video, two widely different views of a scene, two paintings depicting similar content, etc. Whereas each such task is typically addressed with a domain-specific solution, we show that a simple unsupervised approach performs surprisingly well across a range of tasks. Our main insight is that parametric and non-parametric alignment methods have complementary strengths. We propose a two-stage process: first, a feature-based parametric coarse alignment using one or more homographies, followed by non-parametric fine pixel-wise alignment. Coarse alignment is performed using RANSAC on off-the-shelf deep features. Fine alignment is learned in an unsupervised way by a deep network which optimizes a standard structural similarity metric (SSIM) between the two images, plus cycle-consistency. Despite its simplicity, our method shows competitive results on a range of tasks and datasets, including unsupervised optical flow on KITTI, dense correspondences on Hpatches, two-view geometry estimation on YFCC100M, localization on Aachen Day-Night, and, for the first time, fine alignment of artworks on the Brughel dataset. Our code and data are available at http://imagine.enpc.fr/~shenx/RANSAC-Flow/

Xi Shen, Ilaria Pastrolin, Oumayma Bounou et al.

Historical watermark recognition is a highly practical, yet unsolved challenge for archivists and historians. With a large number of well-defined classes, cluttered and noisy samples, different types of representations, both subtle differences between classes and high intra-class variation, historical watermarks are also challenging for pattern recognition. In this paper, overcoming the difficulty of data collection, we present a large public dataset with more than 6k new photographs, allowing for the first time to tackle at scale the scenarios of practical interest for scholars: one-shot instance recognition and cross-domain one-shot instance recognition amongst more than 16k fine-grained classes. We demonstrate that this new dataset is large enough to train modern deep learning approaches, and show that standard methods can be improved considerably by using mid-level deep features. More precisely, we design both a matching score and a feature fine-tuning strategy based on filtering local matches using spatial consistency. This consistency-based approach provides important performance boost compared to strong baselines. Our model achieves 55% top-1 accuracy on our very challenging 16,753-class one-shot cross-domain recognition task, each class described by a single drawing from the classic Briquet catalog. In addition to watermark classification, we show our approach provides promising results on fine-grained sketch-based image retrieval.

Theo Deprelle, Thibault Groueix, Matthew Fisher et al.

We propose to represent shapes as the deformation and combination of learnable elementary 3D structures, which are primitives resulting from training over a collection of shape. We demonstrate that the learned elementary 3D structures lead to clear improvements in 3D shape generation and matching. More precisely, we present two complementary approaches for learning elementary structures: (i) patch deformation learning and (ii) point translation learning. Both approaches can be extended to abstract structures of higher dimensions for improved results. We evaluate our method on two tasks: reconstructing ShapeNet objects and estimating dense correspondences between human scans (FAUST inter challenge). We show 16% improvement over surface deformation approaches for shape reconstruction and outperform FAUST inter challenge state of the art by 6%.

Thibault Groueix, Matthew Fisher, Vladimir G. Kim et al.

We propose a self-supervised approach to deep surface deformation. Given a pair of shapes, our algorithm directly predicts a parametric transformation from one shape to the other respecting correspondences. Our insight is to use cycle-consistency to define a notion of good correspondences in groups of objects and use it as a supervisory signal to train our network. Our method does not rely on a template, assume near isometric deformations or rely on point-correspondence supervision. We demonstrate the efficacy of our approach by using it to transfer segmentation across shapes. We show, on Shapenet, that our approach is competitive with comparable state-of-the-art methods when annotated training data is readily available, but outperforms them by a large margin in the few-shot segmentation scenario.

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.

Yann Labbé, Sergey Zagoruyko, Igor Kalevatykh et al.

We address the problem of visually guided rearrangement planning with many movable objects, i.e., finding a sequence of actions to move a set of objects from an initial arrangement to a desired one, while relying on visual inputs coming from an RGB camera. To do so, we introduce a complete pipeline relying on two key contributions. First, we introduce an efficient and scalable rearrangement planning method, based on a Monte-Carlo Tree Search exploration strategy. We demonstrate that because of its good trade-off between exploration and exploitation our method (i) scales well with the number of objects while (ii) finding solutions which require a smaller number of moves compared to the other state-of-the-art approaches. Note that on the contrary to many approaches, we do not require any buffer space to be available. Second, to precisely localize movable objects in the scene, we develop an integrated approach for robust multi-object workspace state estimation from a single uncalibrated RGB camera using a deep neural network trained only with synthetic data. We validate our multi-object visually guided manipulation pipeline with several experiments on a real UR-5 robotic arm by solving various rearrangement planning instances, requiring only 60 ms to compute the plan to rearrange 25 objects. In addition, we show that our system is insensitive to camera movements and can successfully recover from external perturbations. Supplementary video, source code and pre-trained models are available at https://ylabbe.github.io/rearrangement-planning.

Xi Shen, Alexei A. Efros, Mathieu Aubry

Our goal in this paper is to discover near duplicate patterns in large collections of artworks. This is harder than standard instance mining due to differences in the artistic media (oil, pastel, drawing, etc), and imperfections inherent in the copying process. The key technical insight is to adapt a standard deep feature to this task by fine-tuning it on the specific art collection using self-supervised learning. More specifically, spatial consistency between neighbouring feature matches is used as supervisory fine-tuning signal. The adapted feature leads to more accurate style-invariant matching, and can be used with a standard discovery approach, based on geometric verification, to identify duplicate patterns in the dataset. The approach is evaluated on several different datasets and shows surprisingly good qualitative discovery results. For quantitative evaluation of the method, we annotated 273 near duplicate details in a dataset of 1587 artworks attributed to Jan Brueghel and his workshop. Beyond artwork, we also demonstrate improvement on localization on the Oxford5K photo dataset as well as on historical photograph localization on the Large Time Lags Location (LTLL) dataset.

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.

Othman Sbai, Camille Couprie, Mathieu Aubry

Deep image generation is becoming a tool to enhance artists and designers creativity potential. In this paper, we aim at making the generation process more structured and easier to interact with. Inspired by vector graphics systems, we propose a new deep image reconstruction paradigm where the outputs are composed from simple layers, defined by their color and a vector transparency mask. This presents a number of advantages compared to the commonly used convolutional network architectures. In particular, our layered decomposition allows simple user interaction, for example to update a given mask, or change the color of a selected layer. From a compact code, our architecture also generates vector images with a virtually infinite resolution, the color at each point in an image being a parametric function of its coordinates. We validate the efficiency of our approach by comparing reconstructions with state-of-the-art baselines given similar memory resources on CelebA and ImageNet datasets. Most importantly, we demonstrate several applications of our new image representation obtained in an unsupervised manner, including editing, vectorization and image search.

Thibault Groueix, Matthew Fisher, Vladimir G. Kim et al.

We present a new deep learning approach for matching deformable shapes by introducing {\it Shape Deformation Networks} which jointly encode 3D shapes and correspondences. This is achieved by factoring the surface representation into (i) a template, that parameterizes the surface, and (ii) a learnt global feature vector that parameterizes the transformation of the template into the input surface. By predicting this feature for a new shape, we implicitly predict correspondences between this shape and the template. We show that these correspondences can be improved by an additional step which improves the shape feature by minimizing the Chamfer distance between the input and transformed template. We demonstrate that our simple approach improves on state-of-the-art results on the difficult FAUST-inter challenge, with an average correspondence error of 2.88cm. We show, on the TOSCA dataset, that our method is robust to many types of perturbations, and generalizes to non-human shapes. This robustness allows it to perform well on real unclean, meshes from the the SCAPE dataset.

Thibault Groueix, Matthew Fisher, Vladimir G. Kim et al.

We introduce a method for learning to generate the surface of 3D shapes. Our approach represents a 3D shape as a collection of parametric surface elements and, in contrast to methods generating voxel grids or point clouds, naturally infers a surface representation of the shape. Beyond its novelty, our new shape generation framework, AtlasNet, comes with significant advantages, such as improved precision and generalization capabilities, and the possibility to generate a shape of arbitrary resolution without memory issues. We demonstrate these benefits and compare to strong baselines on the ShapeNet benchmark for two applications: (i) auto-encoding shapes, and (ii) single-view reconstruction from a still image. We also provide results showing its potential for other applications, such as morphing, parametrization, super-resolution, matching, and co-segmentation.

Johanna Delanoy, Mathieu Aubry, Phillip Isola et al.

Sketch-based modeling strives to bring the ease and immediacy of drawing to the 3D world. However, while drawings are easy for humans to create, they are very challenging for computers to interpret due to their sparsity and ambiguity. We propose a data-driven approach that tackles this challenge by learning to reconstruct 3D shapes from one or more drawings. At the core of our approach is a deep convolutional neural network (CNN) that predicts occupancy of a voxel grid from a line drawing. This CNN provides us with an initial 3D reconstruction as soon as the user completes a single drawing of the desired shape. We complement this single-view network with an updater CNN that refines an existing prediction given a new drawing of the shape created from a novel viewpoint. A key advantage of our approach is that we can apply the updater iteratively to fuse information from an arbitrary number of viewpoints, without requiring explicit stroke correspondences between the drawings. We train both CNNs by rendering synthetic contour drawings from hand-modeled shape collections as well as from procedurally-generated abstract shapes. Finally, we integrate our CNNs in a minimal modeling interface that allows users to seamlessly draw an object, rotate it to see its 3D reconstruction, and refine it by re-drawing from another vantage point using the 3D reconstruction as guidance. The main strengths of our approach are its robustness to freehand bitmap drawings, its ability to adapt to different object categories, and the continuum it offers between single-view and multi-view sketch-based modeling.

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.

Tinghui Zhou, Philipp Krähenbühl, Mathieu Aubry et al.

Discriminative deep learning approaches have shown impressive results for problems where human-labeled ground truth is plentiful, but what about tasks where labels are difficult or impossible to obtain? This paper tackles one such problem: establishing dense visual correspondence across different object instances. For this task, although we do not know what the ground-truth is, we know it should be consistent across instances of that category. We exploit this consistency as a supervisory signal to train a convolutional neural network to predict cross-instance correspondences between pairs of images depicting objects of the same category. For each pair of training images we find an appropriate 3D CAD model and render two synthetic views to link in with the pair, establishing a correspondence flow 4-cycle. We use ground-truth synthetic-to-synthetic correspondences, provided by the rendering engine, to train a ConvNet to predict synthetic-to-real, real-to-real and real-to-synthetic correspondences that are cycle-consistent with the ground-truth. At test time, no CAD models are required. We demonstrate that our end-to-end trained ConvNet supervised by cycle-consistency outperforms state-of-the-art pairwise matching methods in correspondence-related tasks.

Francisco Massa, Bryan Russell, Mathieu Aubry

This paper presents an end-to-end convolutional neural network (CNN) for 2D-3D exemplar detection. We demonstrate that the ability to adapt the features of natural images to better align with those of CAD rendered views is critical to the success of our technique. We show that the adaptation can be learned by compositing rendered views of textured object models on natural images. Our approach can be naturally incorporated into a CNN detection pipeline and extends the accuracy and speed benefits from recent advances in deep learning to 2D-3D exemplar detection. We applied our method to two tasks: instance detection, where we evaluated on the IKEA dataset, and object category detection, where we out-perform Aubry et al. for "chair" detection on a subset of the Pascal VOC dataset.

Mathieu Aubry, Bryan Russell

We introduce an approach for analyzing the variation of features generated by convolutional neural networks (CNNs) with respect to scene factors that occur in natural images. Such factors may include object style, 3D viewpoint, color, and scene lighting configuration. Our approach analyzes CNN feature responses corresponding to different scene factors by controlling for them via rendering using a large database of 3D CAD models. The rendered images are presented to a trained CNN and responses for different layers are studied with respect to the input scene factors. We perform a decomposition of the responses based on knowledge of the input scene factors and analyze the resulting components. In particular, we quantify their relative importance in the CNN responses and visualize them using principal component analysis. We show qualitative and quantitative results of our study on three CNNs trained on large image datasets: AlexNet, Places, and Oxford VGG. We observe important differences across the networks and CNN layers for different scene factors and object categories. Finally, we demonstrate that our analysis based on computer-generated imagery translates to the network representation of natural images.