Search

Recovering the geometry of a human head from a single image, while factorizing the materials and illumination, is a severely ill-posed problem that requires prior information to be solved. Methods based on 3D Morphable Models (3DMM), and their combination with differentiable renderers, have shown promising results. However, the expressiveness of 3DMMs is limited, and they typically yield oversmoothed and identity-agnostic 3D shapes limited to the face region. Highly accurate full head reconstructions have recently been obtained with neural fields that parameterize the geometry using multilayer perceptrons. The versatility of these representations has also proved effective for disentangling geometry, materials and lighting. However, these methods require several tens of input images. In this paper, we introduce SIRA, a method which, from a single image, reconstructs human head avatars with high fidelity geometry and factorized lights and surface materials. Our key ingredients are two data-driven statistical models based on neural fields that resolve the ambiguities of single-view 3D surface reconstruction and appearance factorization. Experiments show that SIRA obtains state of the art results in 3D head reconstruction while at the same time it successfully disentangles the global illumination, and the diffuse and specular albedos. Furthermore, our reconstructions are amenable to physically-based appearance editing and head model relighting.

This paper tackles the problem of human motion prediction, consisting in forecasting future body poses from historically observed sequences. State-of-the-art approaches provide good results, however, they rely on deep learning architectures of arbitrary complexity, such as Recurrent Neural Networks (RNN), Transformers or Graph Convolutional Networks (GCN), typically requiring multiple training stages and more than 2 million parameters. In this paper, we show that, after combining with a series of standard practices, such as applying Discrete Cosine Transform (DCT), predicting residual displacement of joints and optimizing velocity as an auxiliary loss, a light-weight network based on multi-layer perceptrons (MLPs) with only 0.14 million parameters can surpass the state-of-the-art performance. An exhaustive evaluation on the Human3.6M, AMASS, and 3DPW datasets shows that our method, named siMLpe, consistently outperforms all other approaches. We hope that our simple method could serve as a strong baseline for the community and allow re-thinking of the human motion prediction problem. The code is publicly available at https://github.com/dulucas/siMLPe.

Automatically producing instructions to modify one's posture could open the door to endless applications, such as personalized coaching and in-home physical therapy. Tackling the reverse problem (i.e., refining a 3D pose based on some natural language feedback) could help for assisted 3D character animation or robot teaching, for instance. Although a few recent works explore the connections between natural language and 3D human pose, none focus on describing 3D body pose differences. In this paper, we tackle the problem of correcting 3D human poses with natural language. To this end, we introduce the Pose- Fix dataset, which consists of several thousand paired 3D poses and their corresponding text feedback, that describe how the source pose needs to be modified to obtain the target pose. We demonstrate the potential of this dataset on two tasks: (1) text-based pose editing, that aims at generating corrected 3D body poses given a query pose and a text modifier; and (2) correctional text generation, where instructions are generated based on the differences between two body poses. The dataset and the code are available at https://europe.naverlabs.com/research/computer-vision/posefix/.

While original Neural Radiance Fields (NeRF) have shown impressive results in modeling the appearance of a scene with compact MLP architectures, they are not able to achieve real-time rendering. This has been recently addressed by either baking the outputs of NeRF into a data structure or arranging trainable parameters in an explicit feature grid. These strategies, however, significantly increase the memory footprint of the model which prevents their deployment on bandwidth-constrained applications. In this paper, we extend the grid-based approach to achieve real-time view synthesis at more than 150 FPS using a lightweight model. Our main contribution is a novel architecture in which the density field of NeRF-based representations is split into N regions and the density is modeled using N different decoders which reuse the same feature grid. This results in a smaller grid where each feature is located in more than one spatial position, forcing them to learn a compact representation that is valid for different parts of the scene. We further reduce the size of the final model by disposing of the features symmetrically on each region, which favors feature pruning after training while also allowing smooth gradient transitions between neighboring voxels. An exhaustive evaluation demonstrates that our method achieves real-time performance and quality metrics on a pair with state-of-the-art with an improvement of more than 2× in the FPS/MB ratio.

3D human pose and shape estimation from monocular images has been an active research area in computer vision. Existing deep learning methods for this task rely on high-resolution input, which however, is not always available in many scenarios such as video surveillance and sports broadcasting. Two common approaches to deal with low-resolution images are applying super-resolution techniques to the input, which may result in unpleasant artifacts, or simply training one model for each resolution, which is impractical in many realistic applications. To address the above issues, this paper proposes a novel algorithm called RSC-Net, which consists of a Resolution-aware network, a Self-supervision loss, and a Contrastive learning scheme. The proposed method is able to learn 3D body pose and shape across different resolutions with one single model. The self-supervision loss enforces scale-consistency of the output, and the contrastive learning scheme enforces scale-consistency of the deep features. We show that both these new losses provide robustness when learning in a weakly-supervised manner. Moreover, we extend the RSC-Net to handle low-resolution videos and apply it to reconstruct textured 3D pedestrians from low-resolution input. Extensive experiments demonstrate that the RSC-Net can achieve consistently better results than the state-of-the-art methods for challenging low-resolution images.

In this work we present a novel attention deep learning model that uses context and human intention for 3D human body motion prediction in handover human-robot tasks. This model uses a multi-head attention architecture which incorporates as inputs the human motion, the robot end effector and the position of the obstacles. The outputs of the model are the predicted motion of the human body and the predicted human intention. We use this model to analyze a handover collaborative task with a robot where the robot is able to predict the future motion of the human and use this information in it¿s planner. Several experiments are performed where human volunteers fill a standard poll to rate different features, taking into account when the robot uses the prediction versus when the robot doesn¿t use the prediction.

As processing power becomes more affordable, com- puter vision tends to push towards controlling complex processes in the industry. Surface inspection, with changing environmental conditions and the usual lack of homogeneity of the inspected parts, makes it a real challenge to overcome even for a skilled specialist. In addition, the scarcity of positive samples and the extremely small size of the defects, makes it even harder to cluster them in different classes. In this work, we propose a novel train- ing strategy tailored to handle these challenges for the problem of image defect segmentation and classification. First, we propose a Context Aggregation Network with different dilation factors, in order to keep as much information as possible from every feature map, especially for the smallest defects. By splitting the loss in classification and segmentation and positively weighing both terms, we accomplish an optimal learning process counteracting possible imbalances in the dataset. Additionally, we introduce a novel guided-crop image augmentation method, which generates new images by cropping real defects from existing images, pasting them in real non-defective ones and finally tweaking their configuration. This augmentation strategically performed, guided by the evolution of each class loss, allows the model to identify better the least common and complicated to identify defects. We validate our solution with the Magnetic Tile and the Severstal Steel Defect Detection dataset, demonstrating that our approach consistently outperforms models such as ResNet-50, DenseNet- 121, HRNet or UPerNet.

We investigated the use of impact sounds generated during exploratory behaviors in a robotic manipulation setup as cues for predicting object surface material and for recognizing individual objects. We collected and make available the YCB-impact sounds dataset which includes over 3,000 impact sounds for the YCB set of everyday objects lying on a table. Impact sounds were generated in three modes: (i) human holding a gripper and hitting, scratching, or dropping the object; (ii) gripper attached to a teleoperated robot hitting the object from the top; (iii) autonomously operated robot hitting the objects from the side with two different speeds. A convolutional neural network is trained from scratch to recognize the object material (steel, aluminium, hard plastic, soft plastic, other plastic, ceramic, wood, paper/cardboard, foam, glass, rubber) from a single impact sound. On the manually collected dataset with more variability in the speed of the action, nearly 60% accuracy for the test set (not presented objects) was achieved. On a robot setup and a stereotypical poking action from top, accuracy of 85% was achieved. This performance drops to 79% if multiple exploratory actions are combined. Individual objects from the set of 75 objects can be recognized with a 79% accuracy. This work demonstrates promising results regarding the possibility of using impact sound for recognition in tasks like single-stream recycling where objects have to be sorted based on their material composition.

Learning controllers that reproduce legged locomotion in nature has been a long-time goal in robotics and computer graphics. While yielding promising results, recent approaches are not yet flexible enough to be applicable to legged systems of different morphologies. This is partly because they often rely on precise motion capture references or elaborate learning environments that ensure the naturality of the emergent locomotion gaits but prevent generalization. This work proposes a generic approach for ensuring realism in locomotion by guiding the learning process with the spring-loaded inverted pendulum model as a reference. Leveraging on the exploration capacities of Reinforcement Learning (RL), we learn a control policy that fills in the information gap between the template model and full-body dynamics required to maintain stable and periodic locomotion. The proposed approach can be applied to robots of different sizes and morphologies and adapted to any RL technique and control architecture. We present experimental results showing that even in a model-free setup and with a simple reactive control architecture, the learned policies can generate realistic and energy-efficient locomotion gaits for a bipedal and a quadrupedal robot. And most importantly, this is achieved without using motion capture, strong constraints in the dynamics or kinematics of the robot, nor prescribing limb coordination. We provide supplemental videos for qualitative analysis of the naturality of the learned gaits.

Neural Radiance Fields (NeRF) has become a popular framework for learning implicit 3D representations and addressing different tasks such as novel-view synthesis or depth-map estimation. However, in downstream applications where decisions need to be made based on automatic predictions, it is critical to leverage the confidence associated with the model estimations. Whereas uncertainty quantification is a long-standing problem in Machine Learning, it has been largely overlooked in the recent NeRF literature. In this context, we propose Stochastic Neural Radiance Fields (S-NeRF), a generalization of standard NeRF that learns a probability distribution over all the possible radiance fields modeling the scene. This distribution allows to quantify the uncertainty associated with the scene information provided by the model. S-NeRF optimization is posed as a Bayesian learning problem that is efficiently addressed using the Variational Inference framework. Exhaustive experiments over benchmark datasets demonstrate that S-NeRF is able to provide more reliable predictions and confidence values than generic approaches previously proposed for uncertainty estimation in other domains.

While there exists many methods for manipulating rigid objects with parallel-jaw grippers, grasping with multi- finger robotic hands remains a quite unexplored research topic. Reasoning and planning collision-free trajectories on the additional degrees of freedom of several fingers represents an important challenge that, so far, involves computationally costly and slow processes. In this work, we present Multi-FinGAN, a fast generative multi-finger grasp sampling method that synthesizes high quality grasps directly from RGB-D images in about a second. We achieve this by training in an end-to-end fashion a coarse-to-fine model composed of a classification network that distinguishes grasp types according to a specific taxonomy and a refinement network that produces refined grasp poses and joint angles. We experimentally validate and benchmark our method against a standard grasp-sampling method on 790 grasps in simulation and 20 grasps on a real Franka Emika Panda. All experimental results using our method show consistent improvements both in terms of grasp quality metrics and grasp success rate. Remarkably, our approach is up to 20-30 times faster than the baseline, a significant improvement that opens the door to feedback-based grasp re-planning and task informative grasping. Code is available at https://irobotics.aalto.fi/multi-fingan/.

We present a novel vehicle detection and tracking system that works solely on 3D LiDAR information. Our approach segments vehicles using a dual-view representation of the 3D LiDAR point cloud on two independently trained convolutional neural networks, one for each view. A bounding box growing algorithm is applied to the fused output of the networks to properly enclose the segmented vehicles. Bounding boxes are grown using a probabilistic method that takes into account also occluded areas. The final vehicle bounding boxes act as observations for a multi-hypothesis tracking system which allows to estimate the position and velocity of the observed vehicles. We thoroughly evaluate our system on the KITTI benchmarks both for detection and tracking separately and show that our dual-branch classifier consistently outperforms previous single-branch approaches, improving or directly competing to other state of the art LiDAR-based methods.

In this paper we introduce SMPLicit, a novel generative model to jointly represent body pose, shape and clothing geometry. In contrast to existing learning-based approaches that require training specific models for each type of garment, SMPLicit can represent in a unified manner different garment topologies (eg from sleeveless tops to hoodies and to open jackets), while controlling other properties like the garment size or tightness/looseness. We show our model to be applicable to a large variety of garments including T-shirts, hoodies, jackets, shorts, pants, skirts, shoes and even hair. The representation flexibility of SMPLicit builds upon an implicit model conditioned with the SMPL human body parameters and a learnable latent space which is semantically interpretable and aligned with the clothing attributes. The proposed model is fully differentiable, allowing for its use into larger end-to-end trainable systems. In the experimental section, we demonstrate SMPLicit can be readily used for fitting 3D scans and for 3D reconstruction in images of dressed people. In both cases we are able to go beyond state of the art, by retrieving complex garment geometries, handling situations with multiple clothing layers and providing a tool for easy outfit editing. To stimulate further research in this direction, we will make our code and model publicly available at http://www.iri.upc.edu/people/ecorona/smplicit/.

Perspective-n-Point-and-Line (PnPL) algorithms aim at fast, accurate, and robust camera localization with respect to a 3D model from 2D-3D feature correspondences, being a major part of modern robotic and AR/VR systems. Current point-based pose estimation methods use only 2D feature detection uncertainties, and the line-based methods do not take uncertainties into account. In our setup, both 3D coordinates and 2D projections of the features are considered uncertain. We propose PnP(L) solvers based on EPnP and DLS for the uncertainty-aware pose estimation. We also modify motion-only bundle adjustment to take 3D uncertainties into account. We perform exhaustive synthetic and real experiments on two different visual odometry datasets. The new PnP(L) methods outperform the state-of-the-art on real data in isolation, showing an increase in mean translation accuracy by 18% on a representative subset of KITTI, while the new uncertain refinement improves pose accuracy for most of the solvers, e.g. decreasing mean translation error for the EPnP by 16% compared to the standard refinement on the same dataset.

This work proposes a human motion prediction model for handover operations. We use in this work, the different phases of the handover operation to improve the human motion predictions. Our attention deep learning based model takes into account the position of the robot’s End Effector (REE) and the phase in the handover operation to predict future human poses. Our model outputs a distribution of possible positions rather than one deterministic position, a key feature in order to allow robots to collaborate with humans. We provide results of the human upper body and the human right hand, also referred as Human End Effector (HEE).The attention deep learning based model has been trained and evaluated with a dataset created using human volunteers and an anthropomorphic robot, simulating handover operations where the robot is the giver and the human the receiver. For each operation, the human skeleton is obtained with an Intel RealSense D435i camera attached inside the robot’s head. The results shown a great improvement of the human’s right hand prediction and 3D body compared with other methods.

Neural rendering techniques combining machine learning with geometric reasoning have arisen as one of the most promising approaches for synthesizing novel views of a scene from a sparse set of images. Among these, stands out the Neural radiance fields (NeRF), which trains a deep network to map 5D input coordinates (representing spatial location and viewing direction) into a volume density and view-dependent emitted radiance. However, despite achieving an unprecedented level of photorealism on the generated images, NeRF is only applicable to static scenes, where the same spatial location can be queried from different images. In this paper we introduce D-NeRF, a method that extends neural radiance fields to a dynamic domain, allowing to reconstruct and render novel images of objects under rigid and non-rigid motions. For this purpose we consider time as an additional input to the system, and split the learning process in two main stages: one that encodes the scene into a canonical space and another that maps this canonical representation into the deformed scene at a particular time. Both mappings are learned using fully-connected networks. Once the networks are trained, D-NeRF can render novel images, controlling both the camera view and the time variable, and thus, the object movement. We demonstrate the effectiveness of our approach on scenes with objects under rigid, articulated and non-rigid motions.

We introduce PhysXNet, a learning-based approach to predict the dynamics of deformable clothes given 3D skeleton motion sequences of humans wearing these clothes. The proposed model is adaptable to a large variety of garments and changing topologies, without need of being retrained. Such simulations are typically carried out by physics engines that require manual human expertise and are subject to computationally intensive computations. PhysXNet, by contrast, is a fully differentiable deep network that at inference is able to estimate the geometry of dense cloth meshes in a matter of milliseconds, and thus, can be readily deployed as a layer of a larger deep learning architecture. This efficiency is achieved thanks to the specific parameterization of the clothes we consider, based on 3D UV maps encoding spatial garment displacements.

In this work we present Sider, a method for high-fidelity detailed 3D face reconstruction from a single image that can be trained in an unsupervised manner. Our approach combines the best from classical statistical models and recent implicit neural representations. The former is used to obtain a coarse shape prior, and the latter provides high-frequency geometric detail, by only optimizing over a photometric loss computed w.r.t. the input image. A thorough quantitative and qualitative evaluation shows that Sider outperforms current state-of-the-art by a significant margin. A limitation of our current approach is that it still cannot handle details like hair or beards and accessories such as glasses. This is because the photometric loss for these regions would require sub-pixel accuracy. In the future, we will explore alternatives for addressing this type of high-frequency details.

Very recently, the Neural Cellular Automata (NCA) has been proposed to simulate the morphogenesis process with deep networks. NCA learns to grow an image starting from a fixed single pixel. In this work, we show that the neural network (NN) architecture of the NCA can be encapsulated in a larger NN. This allows us to propose a new model that encodes a manifold of NCA, each of them capable of generating a distinct image. Therefore, we are effectively learning an embedding space of CA, which shows generalization capabilities. We accomplish this by introducing dynamic convolutions inside an Auto-Encoder architecture, for the first time used to join two different sources of information, the encoding and cell¿s environment information. In biological terms, our approach would play the role of the transcription factors, modulating the mapping of genes into specific proteins that drive cellular differentiation, which occurs right before the morphogenesis. We thoroughly evaluate our approach in a dataset of synthetic emojis and also in real images of CIFAR-10. Our model introduces a general- purpose network, which can be used in a broad range of problems beyond image generation.

Recent literature addressed the monocular 3D pose estimation task very satisfactorily. In these studies, different persons are usually treated as independent pose instances to estimate. However, in many every-day situations, people are interacting, and the pose of an individual depends on the pose of his/her interactees. In this paper, we investigate how to exploit this dependency to enhance current ¿ and possibly future ¿ deep networks for 3D monocular pose estimation. Our pose interacting network, or PI-Net, inputs the initial pose estimates of a variable number of interactees into a recurrent architecture used to refine the pose of the person-of-interest. Evaluating such a method is challenging due to the limited availability of public annotated multi-person 3D human pose datasets. We demonstrate the effectiveness of our method in the MuPoTS dataset, setting the new state-of-the-art on it. Qualitative results on other multi-person datasets (for which 3D pose ground-truth is not available) showcase the proposed PI-Net. PI-Net is implemented in PyTorch and the code will be made available upon acceptance of the paper.

We address the problem of multi-person 3D body pose and shape estimation from a single image. While this problem can be addressed by applying single-person approaches multiple times for the same scene, recent works have shown the advantages of building upon deep architectures that simultaneously reason about all people in the scene in a holistic manner by enforcing, e.g., depth order constraints or minimizing interpenetration among reconstructed bodies. However, existing approaches are still unable to capture the size variability of people caused by the inherent body scale and depth ambiguity. In this work we tackle this challenge by devising a novel optimization scheme that learns the appropriate body scale and relative camera pose, by enforcing the feet of all people to remain on the ground floor. A thorough evaluation on MuPoTS-3D and 3DPW datasets demonstrates that our approach is able to robustly estimate the body translation and shape of multiple people while retrieving their spatial arrangement, consistently improving current state-of-the-art, especially in scenes with people of very different heights.

In this paper we propose a novel approach to estimate dense optical flow from sparse lidar data acquired on an autonomous vehicle. This is intended to be used as a drop-in replacement of any image-based optical flow system when images are not reliable due to e.g. adverse weather conditions or at night. In order to infer high resolution 2D flows from discrete range data we devise a three-block architecture of multiscale filters that combines multiple intermediate objectives, both in the lidar and image domain. To train this network we introduce a dataset with approximately 20K lidar samples of the Kitti dataset which we have augmented with a pseudo ground-truth image-based optical flow computed using FlowNet2. We demonstrate the effectiveness of our approach on Kitti, and show that despite using the low-resolution and sparse measurements of the lidar, we can regress dense optical flow maps which are at par with those estimated with image-based methods.

This paper presents an approach to jointly retrieve camera pose, time-varying 3D shape, and automatic clustering based on motion primitives, from incomplete 2D trajectories in a monocular video. We introduce the concept of order-varying temporal regularization in order to exploit video data, that can be indistinctly applied to the 3D shape evolution as well as to the similarities between images. This results in a union of regularized subspaces which effectively encodes the 3D shape deformation. All parameters are learned via augmented Lagrange multipliers, in a unified and unsupervised manner that does not assume any training data at all. Experimental validation is reported on human motion from sparse to dense shapes, providing more robust and accurate solutions than state-of-the-art approaches in terms of 3D reconstruction, while also obtaining motion grouping results.

This paper introduces an approach to simultaneously estimate 3D shape, camera pose, and object and type of deformation clustering, from partial 2D annotations in a multi-instance collection of images. Furthermore, we can indistinctly process rigid and non-rigid categories. This advances existing work, which only addresses the problem for one single object or, if multiple objects are considered, they are assumed to be clustered a priori. To handle this broader version of the problem, we model object deformation using a formulation based on multiple unions of subspaces, able to span from small rigid motion to complex deformations. The parameters of this model are learned via Augmented Lagrange Multipliers, in a completely unsupervised manner that does not require any training data at all. Extensive validation is provided in a wide variety of synthetic and real scenarios, including rigid and non-rigid categories with small and large deformations. In all cases our approach outperforms state-of-the-art in terms of 3D reconstruction accuracy, while also providing clustering results that allow segmenting the images into object instances and their associated type of deformation (or action the object is performing).

Perception technologies in Autonomous Driving are experiencing their golden age due to the advances in Deep Learning. Yet, most of these systems rely on the semantically rich information of RGB images. Deep Learning solutions applied to the data of other sensors typically mounted on autonomous cars (e.g. lidars or radars) are not explored much. In this paper we propose a novel solution to understand the dynamics of moving vehicles of the scene from only lidar information. The main challenge of this problem stems from the fact that we need to disambiguate the proprio-motion of the “observer” vehicle from that of the external “observed” vehicles. For this purpose, we devise a CNN architecture which at testing time is fed with pairs of consecutive lidar scans. However, in order to properly learn the parameters of this network, during training we introduce a series of so-called pretext tasks which also leverage on image data. These tasks include semantic information about vehicleness and a novel lidar-flow feature which combines standard image-based optical flow with lidar scans. We obtain very promising results and show that including distilled image information only during training, allows improving the inference results of the network at test time, even when image data is no longer used.

We propose a method for predicting the 3D shape of a deformable surface from a single view. By contrast with previous approaches, we do not need a pre-registered template of the surface, and our method is robust to the lack of texture and partial occlusions. At the core of our approach is a geometry-aware deep architecture that tackles the problem as usually done in analytic solutions: first perform 2D detection of the mesh and then estimate a 3D shape that is geometrically consistent with the image. We train this architecture in an end-to-end manner using a large dataset of synthetic renderings of shapes under different levels of deformation, material properties, textures and lighting conditions. We evaluate our approach on a test split of this dataset and available real benchmarks, consistently improving state-of-the-art solutions with a significantly lower computational time.

We present a novel approach for synthesizing photorealistic images of people in arbitrary poses using generative adversarial learning. Given an input image of a person and a desired pose represented by a 2D skeleton, our model renders the image of the same person under the new pose, synthesizing novel views of the parts visible in the input image and hallucinating those that are not seen. This problem has recently been addressed in a supervised manner [16, 35], i.e., during training the ground truth images under the new poses are given to the network. We go beyond these approaches by proposing a fully unsupervised strategy. We tackle this challenging scenario by splitting the problem into two principal subtasks. First, we consider a pose conditioned bidirectional generator that maps back the initially rendered image to the original pose, hence being directly comparable to the input image without the need to resort to any training image. Second, we devise a novel loss function that incorporates content and style terms, and aims at producing images of high perceptual quality. Extensive experiments conducted on the DeepFashion dataset demonstrate that the images rendered by our model are very close in appearance to those obtained by fully supervised approaches.

This paper addresses the problem of simultaneously recovering 3D shape, pose and the elastic model of a deformable object from only 2D point tracks in a monocular video. This is a severely under-constrained problem that has been typically addressed by enforcing the shape or the point trajectories to lie on low-rank dimensional spaces. We show that formulating the problem in terms of a low-rank force space that induces the deformation and introducing the elastic model as an additional unknown, allows for a better physical interpretation of the resulting priors and a more accurate representation of the actual object's behavior. In order to simultaneously estimate force, pose, and the elastic model of the object we use an expectation maximization strategy, where each of these parameters are successively learned by partial M-steps. Once the elastic model is learned, it can be transfered to similar objects to code its 3D deformation. Moreover, our approach can robustly deal with missing data, and encode both rigid and non-rigid points under the same formalism. We thoroughly validate the approach on Mocap and real sequences, showing more accurate 3D reconstructions than state-of-the-art, and additionally providing an estimate of the full elastic model with no a priori information.

Flow-based generative models have highly desirable properties like exact log-likelihood evaluation and exact latent-variable inference, however they are still in their infancy and have not received as much attention as alternative generative models. In this paper, we introduce C-Flow, a novel conditioning scheme that brings normalizing flows to an entirely new scenario with great possibilities for multimodal data modeling. C-Flow is based on a parallel sequence of invertible mappings in which a source flow guides the target flow at every step, enabling fine-grained control over the generation process. We also devise a new strategy to model unordered 3D point clouds that, in combination with the conditioning scheme, makes it possible to address 3D reconstruction from a single image and its inverse problem of rendering an image given a point cloud. We demonstrate our conditioning method to be very adaptable, being also applicable to image manipulation, style transfer and multi-modal image-to-image mapping in a diversity of domains, including RGB images, segmentation maps and edge masks.

The rise of deep learning has brought remarkable progress in estimating hand geometry from images where the hands are part of the scene. This paper focuses on a new problem not explored so far, consisting in predicting how a human would grasp one or several objects, given a single RGB image of these objects. This is a problem with enormous potential in e.g. augmented reality, robotics or prosthetic design. In order to predict feasible grasps, we need to understand the semantic content of the image, its geometric structure and all potential interactions with a hand physical model. To this end, we introduce a generative model that jointly reasons in all these levels and 1) regresses the 3D shape and pose of the objects in the scene; 2) estimates the grasp types; and 3) refines the 51-DoF of a 3D hand model that minimize a graspability loss. To train this model we build the YCB-Affordance dataset, that contains more than 133k images of 21 objects in the YCB-Video dataset. We have annotated these images with more than 28M plausible 3D human grasps according to a 33-class taxonomy. A thorough evaluation in synthetic and real images shows that our model can robustly predict realistic grasps, even in cluttered scenes with multiple objects in close contact.

The problem of predicting human motion given a sequence of past observations is at the core of many applications in robotics and computer vision. Current state-of-the-art formulate this problem as a sequence-to-sequence task, in which a historical of 3D skeletons feeds a Recurrent Neural Network (RNN) that predicts future movements, typically in the order of 1 to 2 seconds. However, one aspect that has been obviated so far, is the fact that human motion is inherently driven by interactions with objects and/or other humans in the environment. In this paper, we explore this scenario using a novel context-aware motion prediction architecture. We use a semantic-graph model where the nodes parameterize the human and objects in the scene and the edges their mutual interactions. These interactions are iteratively learned through a graph attention layer, fed with the past observations, which now include both object and human body motions. Once this semantic graph is learned, we inject it to a standard RNN to predict future movements of the human/s and object/s. We consider two variants of our architecture, either freezing the contextual interactions in the future of updating them. A thorough evaluation in the “Whole-Body Human Motion Database” [29] shows that in both cases, our context-aware networks clearly outperform baselines in which the context information is not considered.

Text Spotting in the wild consists of detecting and recognizing text appearing in images (e.g. signboards, traffic signals or brands in clothing or objects). This is a challenging problem due to the complexity of the context where texts appear (uneven backgrounds, shading, occlusions, perspective distortions, etc.). Only a few approaches try to exploit the relation between text and its surrounding environment to better recognize text in the scene. In this paper, we propose a visual context dataset 1 for Text Spotting in the wild, where the publicly available dataset COCO-text [40] has been extended with information about the scene (such as objects and places appearing in the image) to enable researchers to include semantic relations between texts and scene in their Text Spotting systems, and to offer a common framework for such approaches. For each text in an image, we extract three kinds of context information: objects in the scene, image location label and a textual image description (caption). We use state-of-the-art out-of-the-box available tools to extract this additional information. Since this information has textual form, it can be used to leverage text similarity or semantic relation methods into Text Spotting systems, either as a post-processing or in an end-to-end training strategy.

In this paper we address the problem of extracting vehicle 3D pose from 2D RGB images. An accurate methodology is presented that is capable of locating 3D coordinates of 20 pre-defined semantic vehicle points of interest or keypoints from 2D information. The presented two-step pipeline provides a straightforward way of extracting three-dimensional information from planar images and avoiding also the usage of other sensor that would lead to a more expensive and hard to manage system. The main contribution of this work is the presented dedicated network architectures that are able to locate simultaneously occluded and visible semantic points of interest to convert these 2D points into 3D space in a simple but efficient way. The presented method uses a robust network based on Stack-Hourglass architecture for precise prediction of semantic 2D keypoints from vehicles even if they are occluded. Furthermore, in the second step another dedicated network converts the 2D points into 3D world coordinates and therefore, the 3D pose of the vehicle can be automatically extracted, outperforming state-of-the-art techniques in terms of accuracy.

In this paper we present an approach to reconstruct the 3D shape of multiple deforming objects from a collection of sparse, noisy and possibly incomplete 2D point tracks acquired by a single monocular camera. Additionally, the proposed solution estimates the camera motion and reasons about the spatial segmentation (i.e., identifies each of the deforming objects in every frame) and temporal clustering (i.e., splits the sequence into motion primitive actions). This advances competing work, which mainly tackled the problem for one single object and non-occluded tracks. In order to handle several objects at a time from partial observations, we model point trajectories as a union of spatial and temporal subspaces, and optimize the parameters of both modalities, the non-observed point tracks, the camera motion, and the time-varying 3D shape via augmented Lagrange multipliers. The algorithm is fully unsupervised and does not require any training data at all. We thoroughly validate the method on challenging scenarios with several human subjects performing different activities which involve complex motions and close interaction. We show our approach achieves state-of-the-art 3D reconstruction results, while it also provides spatial and temporal segmentation.

Localization and Mapping is an essential compo-nent to enable Autonomous Vehicles navigation, and requiresan accuracy exceeding that of commercial GPS-based systems.Current odometry and mapping algorithms are able to providethis accurate information. However, the lack of robustness ofthese algorithms against dynamic obstacles and environmentalchanges, even for short time periods, forces the generationof new maps on every session without taking advantage ofpreviously obtained ones. In this paper we propose the useof a deep learning architecture to segmentmovableobjectsfrom 3D LiDAR point clouds in order to obtain longer-lasting3D maps. This will in turn allow for better, faster and moreaccurate re-localization and trajectoy estimation on subsequentdays. We show the effectiveness of our approach in a verydynamic and cluttered scenario, a supermarket parking lot.For that, we record several sequences on different days andcompare localization errors with and without ourmovableobjects segmentation method. Results show that we are able toaccurately re-locate over a filtered map, consistently reducingtrajectory errors between an average of35.1% with respectto a non-filtered map version and of47.9% with respect to astandalone map created on the current session.

We present a system to photo-realistically transfer the clothing of a person in a reference image into another person in an unconstrained image or video. Our architecture is based on a GAN equipped with a physical memory that updates an initially incomplete texture map of the clothes that is progressively completed with the new inferred occluded parts. The system is trained in an unsupervised manner. The results are visually appealing and open the possibility to be used in the future as a quick virtual try-on clothing system.

Recent advances in 3D human shape estimation build upon parametric representations that model very well the shape of the naked body, but are not appropriate to represent the clothing geometry. In this paper, we present an approach to model dressed humans and predict their geometry from single images. We contribute in three fundamental aspects of the problem, namely, a new dataset, a novel shape parameterization algorithm and an end-to-end deep generative network for predicting shape. First, we present 3DPeople, a large-scale synthetic dataset with 2.5 Million photo-realistic images of 80 subjects performing 70 activities and wearing diverse outfits. Besides providing textured 3D meshes for clothes and body, we annotate the dataset with segmentation masks, skeletons, depth, normal maps and optical flow. All this together makes 3DPeople suitable for a plethora of tasks.

We propose a Generative Adversarial Network (GAN) to forecast 3D human motion given a sequence of past 3D skeleton poses. While recent GANs have shown promising results, they can only forecast plausible motion over relatively short periods of time (few hundred milliseconds) and typically ignore the absolute position of the skeleton w.r.t. the camera. Our scheme provides long term predictions (two seconds or more) for both the body pose and its absolute position. Our approach builds upon three main contributions. First, we represent the data using a spatio-temporal tensor of 3D skeleton coordinates which allows formulating the prediction problem as an inpainting one, for which GANs work particularly well. Secondly, we design an architecture to learn the joint distribution of body poses and global motion, capable to hypothesize large chunks of the input 3D tensor with missing data. And finally, we argue that the L2 metric, considered so far by most approaches, fails to capture the actual distribution of long-term human motion. We propose two alternative metrics, based on the distribution of frequencies, that are able to capture more realistic motion patterns. Extensive experiments demonstrate our approach to significantly improve the state of the art, while also handling situations in which past observations are corrupted by occlusions, noise and missing frames.

In this paper, we propose a novel interpretable shape model to encode object nonrigidity. We first use the initial frames of a monocular video to recover a rest shape, used later to compute a dissimilarity measure based on a distance matrix measurement. Spectral analysis is then applied to this matrix to obtain a reduced shape basis, that in contrast to existing approaches, can be physically interpreted. In turn, these precomputed shape bases are used to linearly span the deformation of a wide variety of objects. We introduce the low-rank basis into a sequential approach to recover both camera motion and nonrigid shape from the monocular video, by simply optimizing the weights of the linear combination using bundle adjustment. Since the number of parameters to optimize per frame is relatively small, specially when physical priors are considered, our approach is fast and can potentially run in real time. Validation is done in a wide variety of real-world objects, undergoing both inextensible and extensible deformations. Our approach achieves remarkable robustness to artifacts such as noisy and missing measurements and shows an improved performance to competing methods.

Automatically changing the expression and physical features of a face from an input image is a topic that has been traditionally tackled in a 2D domain. In this paper, we bring this problem to 3D and propose a framework that given an input RGB video of a human face under a neutral expression, initially computes his/her 3D shape and then performs a transfer to a new and potentially non-observed expression. For this purpose, we parameterize the rest shape -obtained from standard factorization approaches over the input video- using a triangular mesh which is further clustered into larger macro-segments. The expression transfer problem is then posed as a direct mapping between this shape and a source shape, such as the blend shapes of an off-the-shelf 3D dataset of human facial expressions. The mapping is resolved to be geometrically consistent between 3D models by requiring points in specific regions to map on semantic equivalent regions. We validate the approach on several synthetic and real examples of input faces that largely differ from the source shapes, yielding very realistic expression transfers even in cases with topology changes, such as a synthetic video sequence of a single-eyed cyclops.

Building upon recent Deep Neural Network architectures, current approaches lying in the intersection of Computer Vision and Natural Language Processing have achieved unprecedented breakthroughs in tasks like automatic captioning or image retrieval. Most of these learning methods, though, rely on large training sets of images associated with human annotations that specifically describe the visual content. In this paper we propose to go a step further and explore the more complex cases where textual descriptions are loosely related to the images. We focus on the particular domain of news articles in which the textual content often expresses connotative and ambiguous relations that are only suggested but not directly inferred from images. We introduce an adaptive CNN architecture that shares most of the structure for multiple tasks including source detection, article illustration and geolocation of articles. Deep Canonical Correlation Analysis is deployed for article illustration, and a new loss function based on Great Circle Distance is proposed for geolocation. Furthermore, we present BreakingNews, a novel dataset with approximately 100K news articles including images, text and captions, and enriched with heterogeneous meta-data (such as GPS coordinates and user comments). We show this dataset to be appropriate to explore all aforementioned problems, for which we provide a baseline performance using various Deep Learning architectures, and different representations of the textual and visual features. We report very promising results and bring to light several limitations of current state-of-the-art in this kind of domain, which we hope will help spur progress in the field.

In this paper we introduce the Boosted Random Ferns (BRFs) to rapidly build discriminative classifiers for learning and detecting object categories. At the core of our approach we use standard random ferns, but we introduce four main innovations that let us bring ferns from an instance to a category level, and still retain efficiency. First, we define binary features on the histogram of oriented gradients-domain (as opposed to intensity-), allowing for a better representation of intra-class variability. Second, both the positions where ferns are evaluated within the sliding window, and the location of the binary features for each fern are not chosen completely at random, but instead we use a boosting strategy to pick the most discriminative combination of them. This is further enhanced by our third contribution, that is to adapt the boosting strategy to enable sharing of binary features among different ferns, yielding high recognition rates at a low computational cost. And finally, we show that training can be performed online, for sequentially arriving images. Overall, the resulting classifier can be very efficiently trained, densely evaluated for all image locations in about 0.1 seconds, and provides detection rates similar to competing approaches that require expensive and significantly slower processing times. We demonstrate the effectiveness of our approach by thorough experimentation in publicly available datasets in which we compare against state-of-the-art, and for tasks of both 2D detection and 3D multi-view estimation.

The most standard approach to resolve the inherent ambiguities of the non-rigid structure from motion problem is using low-rank models that approximate deforming shapes by a linear combination of rigid basis. These models are typically global, i.e., each shape basis contributes equally to all points of the surface. While this approach has been shown effective to represent smooth deformations, its performance degrades for surfaces composed of various regions, each following a different deformation rule. Piecewise methods attempt to capture this type of behavior by locally modeling surface patches, although they subsequently require enforcing global constraints to assemble back the patches. In this paper we propose an approach that combines the best of global and local models: it locally considers low-rank models but, by construction, does not need to impose global constraints to guarantee local patch continuity. We achieve this by a simple expectation maximization strategy that besides learning global shape bases, it locally adapts their contribution to each specific surface region. Furthermore, as a side contribution, in order to split the surface into different local patches, we propose a novel physically-based mesh segmentation approach that obeys an energy criterion. The complete framework is evaluated in both synthetic and real datasets, and shows an improved performance to competing methods.

Finding a product in the fashion world can be a daunting task. Everyday, e-commerce sites are updating with thousands of images and their associated metadata (textual information), deepening the problem, akin to finding a needle in a haystack. In this paper, we leverage both the images and textual meta-data and propose a joint multi-modal embedding that maps both the text and images into a common latent space. Distances in the latent space correspond to similarity between products, allowing us to effectively perform retrieval in this latent space, which is both efficient and accurate. We train this embedding using large-scale real world e-commerce data by both minimizing the similarity between related products and using auxiliary classification networks to that encourage the embedding to have semantic meaning. We compare against existing approaches and show significant improvements in retrieval tasks on a large-scale e-commerce dataset. We also provide an analysis of the different metadata.

Most recent approaches to 3D pose estimation from RGB-D images address the problem in a two-stage pipeline. First, they learn a classifier-typically a random forest-to predict the position of each input pixel on the object surface. These estimates are then used to define an energy function that is minimized w.r.t. the object pose. In this paper, we focus on the first stage of the problem and propose a novel classifier based on a depth-aware Convolutional Neural Network. This classifier is able to learn a scale-adaptive regression model that yields very accurate pixel-level predictions, allowing to finally estimate the pose using a simple RANSAC-based scheme, with no need to optimize complex ad hoc energy functions. Our experiments on publicly available datasets show that our approach achieves remarkable improvements over state-of-the-art methods.

We present an approach to reconstruct the 3D shape of multiple deforming objects from incomplete 2D trajectories acquired by a single camera. Additionally, we simultaneously provide spatial segmentation (i.e., we identify each of the objects in every frame) and temporal clustering (i.e., we split the sequence into primitive actions). This advances existing work, which only tackled the problem for one single object and non-occluded tracks. In order to handle several objects at a time from partial observations, we model point trajectories as a union of spatial and temporal subspaces, and optimize the parameters of both modalities, the non-observed point tracks and the 3D shape via augmented Lagrange multipliers. The algorithm is fully unsupervised and results in a formulation which does not need initialization. We thoroughly validate the method on challenging scenarios with several human subjects performing different activities which involve complex motions and close interaction. We show our approach achieves state-of-the-art 3D reconstruction results, while it also provides spatial and temporal segmentation.

This paper addresses the problem of 3D human pose estimation from a single image. We follow a standard two-step pipeline by first detecting the 2D position of the N body joints, and then using these observations to infer 3D pose. For the first step, we use a recent CNN-based detector. For the second step, most existing approaches perform 2N-to-3N regression of the Cartesian joint coordinates. We show that more precise pose estimates can be obtained by representing both the 2D and 3D human poses using $Ntimes N$ distance matrices, and formulating the problem as a 2D-to-3D distance matrix regression. For learning such a regressor we leverage on simple Neural Network architectures, which by construction, enforce positivity and symmetry of the predicted matrices. The approach has also the advantage to naturally handle missing observations and allowing to hypothesize the position of non-observed joints. Quantitative results on Humaneva and Human3.6M datasets demonstrate consistent performance gains over state-of-the-art. Qualitative evaluation on the images in-the-wild of the LSP dataset, using the regressor learned on Human3.6M, reveals very promising generalization results.

We present an approach to detect the main product in fashion images by exploiting the textual metadata associated with each image. Our approach is based on a Convolutional Neural Network and learns a joint embedding of object proposals and textual metadata to predict the main product in the image. We additionally use several complementary classification and overlap losses in order to improve training stability and performance. Our tests on a large-scale dataset taken from eight e-commerce sites show that our approach outperforms strong baselines and is able to accurately detect the main product in a wide diversity of challenging fashion images.

Exploiting RGB-D data by means of Convolutional Neural Networks (CNNs) is at the core of a number of robotics applications, including object detection, scene semantic segmentation and grasping. Most existing approaches, however, exploit RGB-D data by simply considering depth as an additional input channel for the network. In this paper we show that the performance of deep architectures can be boosted by introducing DaConv, a novel, general-purpose CNN block which exploits depth to learn scale-aware feature representations. We demonstrate the benefits of DaConv on a variety of robotics oriented tasks, involving affordance detection, object coordinate regression and contour detection in RGB-D images. In each of these experiments we show the potential of the proposed block and how it can be readily integrated into existing CNN architectures.

Low textured scenes are well known to be one of the main Achilles heels of geometric computer vision algorithms relying on point correspondences, and in particular for visual SLAM. Yet, there are many environments in which, despite being low textured, one can still reliably estimate line-based geometric primitives, for instance in city and indoor scenes, or in the so-called ``Manhattan worlds'', where structured edges are predominant. In this paper we propose a solution to handle these situations. Specifically, we build upon ORB-SLAM, presumably the current state-of-the-art solution both in terms of accuracy as efficiency, and extend its formulation to simultaneously handle both point and line correspondences. We propose a solution that can even work when most of the points are vanished out from the input images, and, interestingly it can be initialized from solely the detection of line correspondences in three consecutive frames. We thoroughly evaluate our approach and the new initialization strategy on the TUM RGB-D benchmark and demonstrate that the use of lines does not only improve the performance of the original ORB-SLAM solution in poorly textured frames, but also systematically improves it in sequence frames combining points and lines, without compromising the efficiency.