Cristian Sminchisescu

Erik Sandström, Martin R. Oswald, Suryansh Kumar et al.

Many hand-held or mixed reality devices are used with a single sensor for 3D reconstruction, although they often comprise multiple sensors. Multi-sensor depth fusion is able to substantially improve the robustness and accuracy of 3D reconstruction methods, but existing techniques are not robust enough to handle sensors which operate with diverse value ranges as well as noise and outlier statistics. To this end, we introduce SenFuNet, a depth fusion approach that learns sensor-specific noise and outlier statistics and combines the data streams of depth frames from different sensors in an online fashion. Our method fuses multi-sensor depth streams regardless of time synchronization and calibration and generalizes well with little training data. We conduct experiments with various sensor combinations on the real-world CoRBS and Scene3D datasets, as well as the Replica dataset. Experiments demonstrate that our fusion strategy outperforms traditional and recent online depth fusion approaches. In addition, the combination of multiple sensors yields more robust outlier handling and more precise surface reconstruction than the use of a single sensor. The source code and data are available at https://github.com/tfy14esa/SenFuNet.

Erik Gärtner, Luke Metz, Mykhaylo Andriluka et al. · deepmind

We propose a new approach to learned optimization where we represent the computation of an optimizer's update step using a neural network. The parameters of the optimizer are then learned by training on a set of optimization tasks with the objective to perform minimization efficiently. Our innovation is a new neural network architecture, Optimus, for the learned optimizer inspired by the classic BFGS algorithm. As in BFGS, we estimate a preconditioning matrix as a sum of rank-one updates but use a Transformer-based neural network to predict these updates jointly with the step length and direction. In contrast to several recent learned optimization-based approaches, our formulation allows for conditioning across the dimensions of the parameter space of the target problem while remaining applicable to optimization tasks of variable dimensionality without retraining. We demonstrate the advantages of our approach on a benchmark composed of objective functions traditionally used for the evaluation of optimization algorithms, as well as on the real world-task of physics-based visual reconstruction of articulated 3d human motion.

Bharat Lal Bhatnagar, Xianghui Xie, Ilya A. Petrov et al.

Modelling interactions between humans and objects in natural environments is central to many applications including gaming, virtual and mixed reality, as well as human behavior analysis and human-robot collaboration. This challenging operation scenario requires generalization to vast number of objects, scenes, and human actions. Unfortunately, there exist no such dataset. Moreover, this data needs to be acquired in diverse natural environments, which rules out 4D scanners and marker based capture systems. We present BEHAVE dataset, the first full body human- object interaction dataset with multi-view RGBD frames and corresponding 3D SMPL and object fits along with the annotated contacts between them. We record around 15k frames at 5 locations with 8 subjects performing a wide range of interactions with 20 common objects. We use this data to learn a model that can jointly track humans and objects in natural environments with an easy-to-use portable multi-camera setup. Our key insight is to predict correspondences from the human and the object to a statistical body model to obtain human-object contacts during interactions. Our approach can record and track not just the humans and objects but also their interactions, modeled as surface contacts, in 3D. Our code and data can be found at: http://virtualhumans.mpi-inf.mpg.de/behave

Thiemo Alldieck, Mihai Zanfir, Cristian Sminchisescu

We present PHORHUM, a novel, end-to-end trainable, deep neural network methodology for photorealistic 3D human reconstruction given just a monocular RGB image. Our pixel-aligned method estimates detailed 3D geometry and, for the first time, the unshaded surface color together with the scene illumination. Observing that 3D supervision alone is not sufficient for high fidelity color reconstruction, we introduce patch-based rendering losses that enable reliable color reconstruction on visible parts of the human, and detailed and plausible color estimation for the non-visible parts. Moreover, our method specifically addresses methodological and practical limitations of prior work in terms of representing geometry, albedo, and illumination effects, in an end-to-end model where factors can be effectively disentangled. In extensive experiments, we demonstrate the versatility and robustness of our approach. Our state-of-the-art results validate the method qualitatively and for different metrics, for both geometric and color reconstruction.

Nikos Kolotouros, Thiemo Alldieck, Andrei Zanfir et al.

We present DreamHuman, a method to generate realistic animatable 3D human avatar models solely from textual descriptions. Recent text-to-3D methods have made considerable strides in generation, but are still lacking in important aspects. Control and often spatial resolution remain limited, existing methods produce fixed rather than animated 3D human models, and anthropometric consistency for complex structures like people remains a challenge. DreamHuman connects large text-to-image synthesis models, neural radiance fields, and statistical human body models in a novel modeling and optimization framework. This makes it possible to generate dynamic 3D human avatars with high-quality textures and learned, instance-specific, surface deformations. We demonstrate that our method is capable to generate a wide variety of animatable, realistic 3D human models from text. Our 3D models have diverse appearance, clothing, skin tones and body shapes, and significantly outperform both generic text-to-3D approaches and previous text-based 3D avatar generators in visual fidelity. For more results and animations please check our website at https://dream-human.github.io.

Erik Gärtner, Mykhaylo Andriluka, Erwin Coumans et al.

We introduce DiffPhy, a differentiable physics-based model for articulated 3d human motion reconstruction from video. Applications of physics-based reasoning in human motion analysis have so far been limited, both by the complexity of constructing adequate physical models of articulated human motion, and by the formidable challenges of performing stable and efficient inference with physics in the loop. We jointly address such modeling and inference challenges by proposing an approach that combines a physically plausible body representation with anatomical joint limits, a differentiable physics simulator, and optimization techniques that ensure good performance and robustness to suboptimal local optima. In contrast to several recent methods, our approach readily supports full-body contact including interactions with objects in the scene. Most importantly, our model connects end-to-end with images, thus supporting direct gradient-based physics optimization by means of image-based loss functions. We validate the model by demonstrating that it can accurately reconstruct physically plausible 3d human motion from monocular video, both on public benchmarks with available 3d ground-truth, and on videos from the internet.

Ivan Grishchenko, Valentin Bazarevsky, Andrei Zanfir et al.

We present BlazePose GHUM Holistic, a lightweight neural network pipeline for 3D human body landmarks and pose estimation, specifically tailored to real-time on-device inference. BlazePose GHUM Holistic enables motion capture from a single RGB image including avatar control, fitness tracking and AR/VR effects. Our main contributions include i) a novel method for 3D ground truth data acquisition, ii) updated 3D body tracking with additional hand landmarks and iii) full body pose estimation from a monocular image.

Erik Gärtner, Mykhaylo Andriluka, Hongyi Xu et al.

We focus on the task of estimating a physically plausible articulated human motion from monocular video. Existing approaches that do not consider physics often produce temporally inconsistent output with motion artifacts, while state-of-the-art physics-based approaches have either been shown to work only in controlled laboratory conditions or consider simplified body-ground contact limited to feet. This paper explores how these shortcomings can be addressed by directly incorporating a fully-featured physics engine into the pose estimation process. Given an uncontrolled, real-world scene as input, our approach estimates the ground-plane location and the dimensions of the physical body model. It then recovers the physical motion by performing trajectory optimization. The advantage of our formulation is that it readily generalizes to a variety of scenes that might have diverse ground properties and supports any form of self-contact and contact between the articulated body and scene geometry. We show that our approach achieves competitive results with respect to existing physics-based methods on the Human3.6M benchmark, while being directly applicable without re-training to more complex dynamic motions from the AIST benchmark and to uncontrolled internet videos.

Andrei Zanfir, Mihai Zanfir, Alexander Gorban et al.

Autonomous driving is an exciting new industry, posing important research questions. Within the perception module, 3D human pose estimation is an emerging technology, which can enable the autonomous vehicle to perceive and understand the subtle and complex behaviors of pedestrians. While hardware systems and sensors have dramatically improved over the decades -- with cars potentially boasting complex LiDAR and vision systems and with a growing expansion of the available body of dedicated datasets for this newly available information -- not much work has been done to harness these novel signals for the core problem of 3D human pose estimation. Our method, which we coin HUM3DIL (HUMan 3D from Images and LiDAR), efficiently makes use of these complementary signals, in a semi-supervised fashion and outperforms existing methods with a large margin. It is a fast and compact model for onboard deployment. Specifically, we embed LiDAR points into pixel-aligned multi-modal features, which we pass through a sequence of Transformer refinement stages. Quantitative experiments on the Waymo Open Dataset support these claims, where we achieve state-of-the-art results on the task of 3D pose estimation.

Enric Corona, Mihai Zanfir, Thiemo Alldieck et al.

We introduce Structured 3D Features, a model based on a novel implicit 3D representation that pools pixel-aligned image features onto dense 3D points sampled from a parametric, statistical human mesh surface. The 3D points have associated semantics and can move freely in 3D space. This allows for optimal coverage of the person of interest, beyond just the body shape, which in turn, additionally helps modeling accessories, hair, and loose clothing. Owing to this, we present a complete 3D transformer-based attention framework which, given a single image of a person in an unconstrained pose, generates an animatable 3D reconstruction with albedo and illumination decomposition, as a result of a single end-to-end model, trained semi-supervised, and with no additional postprocessing. We show that our S3F model surpasses the previous state-of-the-art on various tasks, including monocular 3D reconstruction, as well as albedo and shading estimation. Moreover, we show that the proposed methodology allows novel view synthesis, relighting, and re-posing the reconstruction, and can naturally be extended to handle multiple input images (e.g. different views of a person, or the same view, in different poses, in video). Finally, we demonstrate the editing capabilities of our model for 3D virtual try-on applications.

Mihai Fieraru, Mihai Zanfir, Elisabeta Oneata et al.

Understanding 3d human interactions is fundamental for fine-grained scene analysis and behavioural modeling. However, most of the existing models predict incorrect, lifeless 3d estimates, that miss the subtle human contact aspects--the essence of the event--and are of little use for detailed behavioral understanding. This paper addresses such issues with several contributions: (1) we introduce models for interaction signature estimation (ISP) encompassing contact detection, segmentation, and 3d contact signature prediction; (2) we show how such components can be leveraged to ensure contact consistency during 3d reconstruction; (3) we construct several large datasets for learning and evaluating 3d contact prediction and reconstruction methods; specifically, we introduce CHI3D, a lab-based accurate 3d motion capture dataset with 631 sequences containing $2,525$ contact events, $728,664$ ground truth 3d poses, as well as FlickrCI3D, a dataset of $11,216$ images, with $14,081$ processed pairs of people, and $81,233$ facet-level surface correspondences. Finally, (4) we propose methodology for recovering the ground-truth pose and shape of interacting people in a controlled setup and (5) annotate all 3d interaction motions in CHI3D with textual descriptions. Motion data in multiple formats (GHUM and SMPLX parameters, Human3.6m 3d joints) is made available for research purposes at \url{https://ci3d.imar.ro}, together with an evaluation server and a public benchmark.

Mihai Zanfir, Thiemo Alldieck, Cristian Sminchisescu

We present PhoMoH, a neural network methodology to construct generative models of photo-realistic 3D geometry and appearance of human heads including hair, beards, an oral cavity, and clothing. In contrast to prior work, PhoMoH models the human head using neural fields, thus supporting complex topology. Instead of learning a head model from scratch, we propose to augment an existing expressive head model with new features. Concretely, we learn a highly detailed geometry network layered on top of a mid-resolution head model together with a detailed, local geometry-aware, and disentangled color field. Our proposed architecture allows us to learn photo-realistic human head models from relatively little data. The learned generative geometry and appearance networks can be sampled individually and enable the creation of diverse and realistic human heads. Extensive experiments validate our method qualitatively and across different metrics.

Ivan Grishchenko, Geng Yan, Eduard Gabriel Bazavan et al.

We present Blendshapes GHUM, an on-device ML pipeline that predicts 52 facial blendshape coefficients at 30+ FPS on modern mobile phones, from a single monocular RGB image and enables facial motion capture applications like virtual avatars. Our main contributions are: i) an annotation-free offline method for obtaining blendshape coefficients from real-world human scans, ii) a lightweight real-time model that predicts blendshape coefficients based on facial landmarks.

Letian Wang, Andrei Zanfir, Eduard Gabriel Bazavan et al.

We present THFM, a unified video foundation model for human-centric perception that jointly addresses dense tasks (depth, normals, segmentation, dense pose) and sparse tasks (2d/3d keypoint estimation) within a single architecture. THFM is derived from a pretrained text-to-video diffusion model, repurposed as a single-forward-pass perception model and augmented with learnable tokens for sparse predictions. Modulated by the text prompt, our single unified model is capable of performing various perception tasks. Crucially, our model is on-par or surpassing state-of-the-art specialized models on a variety of benchmarks despite being trained exclusively on synthetic data (i.e.~without training on real-world or benchmark specific data). We further highlight intriguing emergent properties of our model, which we attribute to the underlying diffusion-based video representation. For example, our model trained on videos with a single human in the scene generalizes to multiple humans and other object classes such as anthropomorphic characters and animals -- a capability that hasn't been demonstrated in the past.

Henning Petzka, Ted Kronvall, Cristian Sminchisescu

Interpolations in the latent space of deep generative models is one of the standard tools to synthesize semantically meaningful mixtures of generated samples. As the generator function is non-linear, commonly used linear interpolations in the latent space do not yield the shortest paths in the sample space, resulting in non-smooth interpolations. Recent work has therefore equipped the latent space with a suitable metric to enforce shortest paths on the manifold of generated samples. These are often, however, susceptible of veering away from the manifold of real samples, resulting in smooth but unrealistic generation that requires an additional method to assess the sample quality along paths. Generative Adversarial Networks (GANs), by construction, measure the sample quality using its discriminator network. In this paper, we establish that the discriminator can be used effectively to avoid regions of low sample quality along shortest paths. By reusing the discriminator network to modify the metric on the latent space, we propose a lightweight solution for improved interpolations in pre-trained GANs.

Eduard Gabriel Bazavan, Andrei Zanfir, Thiemo Alldieck et al.

We present \emph{SPHEAR}, an accurate, differentiable parametric statistical 3D human head model, enabled by a novel 3D registration method based on spherical embeddings. We shift the paradigm away from the classical Non-Rigid Registration methods, which operate under various surface priors, increasing reconstruction fidelity and minimizing required human intervention. Additionally, SPHEAR is a \emph{complete} model that allows not only to sample diverse synthetic head shapes and facial expressions, but also gaze directions, high-resolution color textures, surface normal maps, and hair cuts represented in detail, as strands. SPHEAR can be used for automatic realistic visual data generation, semantic annotation, and general reconstruction tasks. Compared to state-of-the-art approaches, our components are fast and memory efficient, and experiments support the validity of our design choices and the accuracy of registration, reconstruction and generation techniques.

Kartik Gupta, Amir Rahimi, Thalaiyasingam Ajanthan et al.

Calibrating neural networks is of utmost importance when employing them in safety-critical applications where the downstream decision making depends on the predicted probabilities. Measuring calibration error amounts to comparing two empirical distributions. In this work, we introduce a binning-free calibration measure inspired by the classical Kolmogorov-Smirnov (KS) statistical test in which the main idea is to compare the respective cumulative probability distributions. From this, by approximating the empirical cumulative distribution using a differentiable function via splines, we obtain a recalibration function, which maps the network outputs to actual (calibrated) class assignment probabilities. The spine-fitting is performed using a held-out calibration set and the obtained recalibration function is evaluated on an unseen test set. We tested our method against existing calibration approaches on various image classification datasets and our spline-based recalibration approach consistently outperforms existing methods on KS error as well as other commonly used calibration measures. Our Code is available at https://github.com/kartikgupta-at-anu/spline-calibration.

Enric Corona, Andrei Zanfir, Eduard Gabriel Bazavan et al.

We propose VLOGGER, a method for audio-driven human video generation from a single input image of a person, which builds on the success of recent generative diffusion models. Our method consists of 1) a stochastic human-to-3d-motion diffusion model, and 2) a novel diffusion-based architecture that augments text-to-image models with both spatial and temporal controls. This supports the generation of high quality video of variable length, easily controllable through high-level representations of human faces and bodies. In contrast to previous work, our method does not require training for each person, does not rely on face detection and cropping, generates the complete image (not just the face or the lips), and considers a broad spectrum of scenarios (e.g. visible torso or diverse subject identities) that are critical to correctly synthesize humans who communicate. We also curate MENTOR, a new and diverse dataset with 3d pose and expression annotations, one order of magnitude larger than previous ones (800,000 identities) and with dynamic gestures, on which we train and ablate our main technical contributions. VLOGGER outperforms state-of-the-art methods in three public benchmarks, considering image quality, identity preservation and temporal consistency while also generating upper-body gestures. We analyze the performance of VLOGGER with respect to multiple diversity metrics, showing that our architectural choices and the use of MENTOR benefit training a fair and unbiased model at scale. Finally we show applications in video editing and personalization.

Thiemo Alldieck, Nikos Kolotouros, Cristian Sminchisescu

Score Distillation Sampling (SDS) is a recent but already widely popular method that relies on an image diffusion model to control optimization problems using text prompts. In this paper, we conduct an in-depth analysis of the SDS loss function, identify an inherent problem with its formulation, and propose a surprisingly easy but effective fix. Specifically, we decompose the loss into different factors and isolate the component responsible for noisy gradients. In the original formulation, high text guidance is used to account for the noise, leading to unwanted side effects such as oversaturation or repeated detail. Instead, we train a shallow network mimicking the timestep-dependent frequency bias of the image diffusion model in order to effectively factor it out. We demonstrate the versatility and the effectiveness of our novel loss formulation through qualitative and quantitative experiments, including optimization-based image synthesis and editing, zero-shot image translation network training, and text-to-3D synthesis.

Akash Sengupta, Thiemo Alldieck, Nikos Kolotouros et al.

We present DiffHuman, a probabilistic method for photorealistic 3D human reconstruction from a single RGB image. Despite the ill-posed nature of this problem, most methods are deterministic and output a single solution, often resulting in a lack of geometric detail and blurriness in unseen or uncertain regions. In contrast, DiffHuman predicts a probability distribution over 3D reconstructions conditioned on an input 2D image, which allows us to sample multiple detailed 3D avatars that are consistent with the image. DiffHuman is implemented as a conditional diffusion model that denoises pixel-aligned 2D observations of an underlying 3D shape representation. During inference, we may sample 3D avatars by iteratively denoising 2D renders of the predicted 3D representation. Furthermore, we introduce a generator neural network that approximates rendering with considerably reduced runtime (55x speed up), resulting in a novel dual-branch diffusion framework. Our experiments show that DiffHuman can produce diverse and detailed reconstructions for the parts of the person that are unseen or uncertain in the input image, while remaining competitive with the state-of-the-art when reconstructing visible surfaces.

Mykhaylo Andriluka, Baruch Tabanpour, C. Daniel Freeman et al.

We propose a novel neural network approach, LARP (Learned Articulated Rigid body Physics), to model the dynamics of articulated human motion with contact. Our goal is to develop a faster and more convenient methodological alternative to traditional physics simulators for use in computer vision tasks such as human motion reconstruction from video. To that end we introduce a training procedure and model components that support the construction of a recurrent neural architecture to accurately simulate articulated rigid body dynamics. Our neural architecture supports features typically found in traditional physics simulators, such as modeling of joint motors, variable dimensions of body parts, contact between body parts and objects, and is an order of magnitude faster than traditional systems when multiple simulations are run in parallel. To demonstrate the value of LARP we use it as a drop-in replacement for a state of the art classical non-differentiable simulator in an existing video-based reconstruction framework and show comparative or better 3D human pose reconstruction accuracy.

Teodor Alexandru Szente, James Harrison, Mihai Zanfir et al.

Fractional gradient descent has been studied extensively, with a focus on its ability to extend traditional gradient descent methods by incorporating fractional-order derivatives. This approach allows for more flexibility in navigating complex optimization landscapes and offers advantages in certain types of problems, particularly those involving non-linearities and chaotic dynamics. Yet, the challenge of fine-tuning the fractional order parameters remains unsolved. In this work, we demonstrate that it is possible to train a neural network to predict the order of the gradient effectively.

Nikos Kolotouros, Thiemo Alldieck, Enric Corona et al.

We present AvatarPopUp, a method for fast, high quality 3D human avatar generation from different input modalities, such as images and text prompts and with control over the generated pose and shape. The common theme is the use of diffusion-based image generation networks that are specialized for each particular task, followed by a 3D lifting network. We purposefully decouple the generation from the 3D modeling which allow us to leverage powerful image synthesis priors, trained on billions of text-image pairs. We fine-tune latent diffusion networks with additional image conditioning for image generation and back-view prediction, and to support qualitatively different multiple 3D hypotheses. Our partial fine-tuning approach allows to adapt the networks for each task without inducing catastrophic forgetting. In our experiments, we demonstrate that our method produces accurate, high-quality 3D avatars with diverse appearance that respect the multimodal text, image, and body control signals. Our approach can produce a 3D model in as few as 2 seconds, a four orders of magnitude speedup wrt the vast majority of existing methods, most of which solve only a subset of our tasks, and with fewer controls. AvatarPopUp enables applications that require the controlled 3D generation of human avatars at scale. The project website can be found at https://www.nikoskolot.com/avatarpopup/.

Maria Priisalu, Ted Kronvall, Cristian Sminchisescu

Human pose forecasting is the task of predicting articulated human motion given past human motion. There exists a number of popular benchmarks that evaluate an array of different models performing human pose forecasting. These benchmarks do not reflect that a human interacting system, such as a delivery robot, observes and plans for the motion of the same individual over an extended period of time. Every individual has unique and distinct movement patterns. This is however not reflected in existing benchmarks that evaluate a model's ability to predict an average human's motion rather than a particular individual's. We reformulate the human motion forecasting problem and present a model-agnostic personalization method. Motion forecasting personalization can be performed efficiently online by utilizing a low-parametric time-series analysis model that personalizes neural network pose predictions.

David Nilsson, Aleksis Pirinen, Erik Gärtner et al.

We study lifelong visual perception in an embodied setup, where we develop new models and compare various agents that navigate in buildings and occasionally request annotations which, in turn, are used to refine their visual perception capabilities. The purpose of the agents is to recognize objects and other semantic classes in the whole building at the end of a process that combines exploration and active visual learning. As we study this task in a lifelong learning context, the agents should use knowledge gained in earlier visited environments in order to guide their exploration and active learning strategy in successively visited buildings. We use the semantic segmentation performance as a proxy for general visual perception and study this novel task for several exploration and annotation methods, ranging from frontier exploration baselines which use heuristic active learning, to a fully learnable approach. For the latter, we introduce a deep reinforcement learning (RL) based agent which jointly learns both navigation and active learning. A point goal navigation formulation, coupled with a global planner which supplies goals, is integrated into the RL model in order to provide further incentives for systematic exploration of novel scenes. By performing extensive experiments on the Matterport3D dataset, we show how the proposed agents can utilize knowledge from previously explored scenes when exploring new ones, e.g. through less granular exploration and less frequent requests for annotations. The results also suggest that a learning-based agent is able to use its prior visual knowledge more effectively than heuristic alternatives.

Eduard Gabriel Bazavan, Andrei Zanfir, Mihai Zanfir et al.

Advances in the state of the art for 3d human sensing are currently limited by the lack of visual datasets with 3d ground truth, including multiple people, in motion, operating in real-world environments, with complex illumination or occlusion, and potentially observed by a moving camera. Sophisticated scene understanding would require estimating human pose and shape as well as gestures, towards representations that ultimately combine useful metric and behavioral signals with free-viewpoint photo-realistic visualisation capabilities. To sustain progress, we build a large-scale photo-realistic dataset, Human-SPACE (HSPACE), of animated humans placed in complex synthetic indoor and outdoor environments. We combine a hundred diverse individuals of varying ages, gender, proportions, and ethnicity, with hundreds of motions and scenes, as well as parametric variations in body shape (for a total of 1,600 different humans), in order to generate an initial dataset of over 1 million frames. Human animations are obtained by fitting an expressive human body model, GHUM, to single scans of people, followed by novel re-targeting and positioning procedures that support the realistic animation of dressed humans, statistical variation of body proportions, and jointly consistent scene placement of multiple moving people. Assets are generated automatically, at scale, and are compatible with existing real time rendering and game engines. The dataset with evaluation server will be made available for research. Our large-scale analysis of the impact of synthetic data, in connection with real data and weak supervision, underlines the considerable potential for continuing quality improvements and limiting the sim-to-real gap, in this practical setting, in connection with increased model capacity.

Hongyi Xu, Thiemo Alldieck, Cristian Sminchisescu

We present neural radiance fields for rendering and temporal (4D) reconstruction of humans in motion (H-NeRF), as captured by a sparse set of cameras or even from a monocular video. Our approach combines ideas from neural scene representation, novel-view synthesis, and implicit statistical geometric human representations, coupled using novel loss functions. Instead of learning a radiance field with a uniform occupancy prior, we constrain it by a structured implicit human body model, represented using signed distance functions. This allows us to robustly fuse information from sparse views and generalize well beyond the poses or views observed in training. Moreover, we apply geometric constraints to co-learn the structure of the observed subject -- including both body and clothing -- and to regularize the radiance field to geometrically plausible solutions. Extensive experiments on multiple datasets demonstrate the robustness and the accuracy of our approach, its generalization capabilities significantly outside a small training set of poses and views, and statistical extrapolation beyond the observed shape.

Thiemo Alldieck, Hongyi Xu, Cristian Sminchisescu

We present imGHUM, the first holistic generative model of 3D human shape and articulated pose, represented as a signed distance function. In contrast to prior work, we model the full human body implicitly as a function zero-level-set and without the use of an explicit template mesh. We propose a novel network architecture and a learning paradigm, which make it possible to learn a detailed implicit generative model of human pose, shape, and semantics, on par with state-of-the-art mesh-based models. Our model features desired detail for human models, such as articulated pose including hand motion and facial expressions, a broad spectrum of shape variations, and can be queried at arbitrary resolutions and spatial locations. Additionally, our model has attached spatial semantics making it straightforward to establish correspondences between different shape instances, thus enabling applications that are difficult to tackle using classical implicit representations. In extensive experiments, we demonstrate the model accuracy and its applicability to current research problems.

Davide Menini, Suryansh Kumar, Martin R. Oswald et al.

This paper presents a real-time online vision framework to jointly recover an indoor scene's 3D structure and semantic label. Given noisy depth maps, a camera trajectory, and 2D semantic labels at train time, the proposed deep neural network based approach learns to fuse the depth over frames with suitable semantic labels in the scene space. Our approach exploits the joint volumetric representation of the depth and semantics in the scene feature space to solve this task. For a compelling online fusion of the semantic labels and geometry in real-time, we introduce an efficient vortex pooling block while dropping the use of routing network in online depth fusion to preserve high-frequency surface details. We show that the context information provided by the semantics of the scene helps the depth fusion network learn noise-resistant features. Not only that, it helps overcome the shortcomings of the current online depth fusion method in dealing with thin object structures, thickening artifacts, and false surfaces. Experimental evaluation on the Replica dataset shows that our approach can perform depth fusion at 37 and 10 frames per second with an average reconstruction F-score of 88% and 91%, respectively, depending on the depth map resolution. Moreover, our model shows an average IoU score of 0.515 on the ScanNet 3D semantic benchmark leaderboard.

Pei Sun, Weiyue Wang, Yuning Chai et al.

The detection of 3D objects from LiDAR data is a critical component in most autonomous driving systems. Safe, high speed driving needs larger detection ranges, which are enabled by new LiDARs. These larger detection ranges require more efficient and accurate detection models. Towards this goal, we propose Range Sparse Net (RSN), a simple, efficient, and accurate 3D object detector in order to tackle real time 3D object detection in this extended detection regime. RSN predicts foreground points from range images and applies sparse convolutions on the selected foreground points to detect objects. The lightweight 2D convolutions on dense range images results in significantly fewer selected foreground points, thus enabling the later sparse convolutions in RSN to efficiently operate. Combining features from the range image further enhance detection accuracy. RSN runs at more than 60 frames per second on a 150m x 150m detection region on Waymo Open Dataset (WOD) while being more accurate than previously published detectors. As of 11/2020, RSN is ranked first in the WOD leaderboard based on the APH/LEVEL 1 metrics for LiDAR-based pedestrian and vehicle detection, while being several times faster than alternatives.

Mihai Zanfir, Andrei Zanfir, Eduard Gabriel Bazavan et al.

We present THUNDR, a transformer-based deep neural network methodology to reconstruct the 3d pose and shape of people, given monocular RGB images. Key to our methodology is an intermediate 3d marker representation, where we aim to combine the predictive power of model-free-output architectures and the regularizing, anthropometrically-preserving properties of a statistical human surface model like GHUM -- a recently introduced, expressive full body statistical 3d human model, trained end-to-end. Our novel transformer-based prediction pipeline can focus on image regions relevant to the task, supports self-supervised regimes, and ensures that solutions are consistent with human anthropometry. We show state-of-the-art results on Human3.6M and 3DPW, for both the fully-supervised and the self-supervised models, for the task of inferring 3d human shape, joint positions, and global translation. Moreover, we observe very solid 3d reconstruction performance for difficult human poses collected in the wild.

Jack Valmadre, Alex Bewley, Jonathan Huang et al.

This paper introduces temporally local metrics for Multi-Object Tracking. These metrics are obtained by restricting existing metrics based on track matching to a finite temporal horizon, and provide new insight into the ability of trackers to maintain identity over time. Moreover, the horizon parameter offers a novel, meaningful mechanism by which to define the relative importance of detection and association, a common dilemma in applications where imperfect association is tolerable. It is shown that the historical Average Tracking Accuracy (ATA) metric exhibits superior sensitivity to association, enabling its proposed local variant, ALTA, to capture a wide range of characteristics. In particular, ALTA is better equipped to identify advances in association independent of detection. The paper further presents an error decomposition for ATA that reveals the impact of four distinct error types and is equally applicable to ALTA. The diagnostic capabilities of ALTA are demonstrated on the MOT 2017 and Waymo Open Dataset benchmarks.

Mihai Fieraru, Mihai Zanfir, Elisabeta Oneata et al.

Monocular estimation of three dimensional human self-contact is fundamental for detailed scene analysis including body language understanding and behaviour modeling. Existing 3d reconstruction methods do not focus on body regions in self-contact and consequently recover configurations that are either far from each other or self-intersecting, when they should just touch. This leads to perceptually incorrect estimates and limits impact in those very fine-grained analysis domains where detailed 3d models are expected to play an important role. To address such challenges we detect self-contact and design 3d losses to explicitly enforce it. Specifically, we develop a model for Self-Contact Prediction (SCP), that estimates the body surface signature of self-contact, leveraging the localization of self-contact in the image, during both training and inference. We collect two large datasets to support learning and evaluation: (1) HumanSC3D, an accurate 3d motion capture repository containing $1,032$ sequences with $5,058$ contact events and $1,246,487$ ground truth 3d poses synchronized with images collected from multiple views, and (2) FlickrSC3D, a repository of $3,969$ images, containing $25,297$ surface-to-surface correspondences with annotated image spatial support. We also illustrate how more expressive 3d reconstructions can be recovered under self-contact signature constraints and present monocular detection of face-touch as one of the multiple applications made possible by more accurate self-contact models.

David Nilsson, Aleksis Pirinen, Erik Gärtner et al.

We study the task of embodied visual active learning, where an agent is set to explore a 3d environment with the goal to acquire visual scene understanding by actively selecting views for which to request annotation. While accurate on some benchmarks, today's deep visual recognition pipelines tend to not generalize well in certain real-world scenarios, or for unusual viewpoints. Robotic perception, in turn, requires the capability to refine the recognition capabilities for the conditions where the mobile system operates, including cluttered indoor environments or poor illumination. This motivates the proposed task, where an agent is placed in a novel environment with the objective of improving its visual recognition capability. To study embodied visual active learning, we develop a battery of agents - both learnt and pre-specified - and with different levels of knowledge of the environment. The agents are equipped with a semantic segmentation network and seek to acquire informative views, move and explore in order to propagate annotations in the neighbourhood of those views, then refine the underlying segmentation network by online retraining. The trainable method uses deep reinforcement learning with a reward function that balances two competing objectives: task performance, represented as visual recognition accuracy, which requires exploring the environment, and the necessary amount of annotated data requested during active exploration. We extensively evaluate the proposed models using the photorealistic Matterport3D simulator and show that a fully learnt method outperforms comparable pre-specified counterparts, even when requesting fewer annotations.

Bharat Lal Bhatnagar, Cristian Sminchisescu, Christian Theobalt et al.

We address the problem of fitting 3D human models to 3D scans of dressed humans. Classical methods optimize both the data-to-model correspondences and the human model parameters (pose and shape), but are reliable only when initialized close to the solution. Some methods initialize the optimization based on fully supervised correspondence predictors, which is not differentiable end-to-end, and can only process a single scan at a time. Our main contribution is LoopReg, an end-to-end learning framework to register a corpus of scans to a common 3D human model. The key idea is to create a self-supervised loop. A backward map, parameterized by a Neural Network, predicts the correspondence from every scan point to the surface of the human model. A forward map, parameterized by a human model, transforms the corresponding points back to the scan based on the model parameters (pose and shape), thus closing the loop. Formulating this closed loop is not straightforward because it is not trivial to force the output of the NN to be on the surface of the human model - outside this surface the human model is not even defined. To this end, we propose two key innovations. First, we define the canonical surface implicitly as the zero level set of a distance field in R3, which in contrast to morecommon UV parameterizations, does not require cutting the surface, does not have discontinuities, and does not induce distortion. Second, we diffuse the human model to the 3D domain R3. This allows to map the NN predictions forward,even when they slightly deviate from the zero level set. Results demonstrate that we can train LoopRegmainly self-supervised - following a supervised warm-start, the model becomes increasingly more accurate as additional unlabelled raw scans are processed. Our code and pre-trained models can be downloaded for research.

Andrei Zanfir, Eduard Gabriel Bazavan, Mihai Zanfir et al.

We present deep neural network methodology to reconstruct the 3d pose and shape of people, given an input RGB image. We rely on a recently introduced, expressivefull body statistical 3d human model, GHUM, trained end-to-end, and learn to reconstruct its pose and shape state in a self-supervised regime. Central to our methodology, is a learning to learn and optimize approach, referred to as HUmanNeural Descent (HUND), which avoids both second-order differentiation when training the model parameters,and expensive state gradient descent in order to accurately minimize a semantic differentiable rendering loss at test time. Instead, we rely on novel recurrent stages to update the pose and shape parameters such that not only losses are minimized effectively, but the process is meta-regularized in order to ensure end-progress. HUND's symmetry between training and testing makes it the first 3d human sensing architecture to natively support different operating regimes including self-supervised ones. In diverse tests, we show that HUND achieves very competitive results in datasets like H3.6M and 3DPW, aswell as good quality 3d reconstructions for complex imagery collected in-the-wild.

Bharat Lal Bhatnagar, Cristian Sminchisescu, Christian Theobalt et al.

Implicit functions represented as deep learning approximations are powerful for reconstructing 3D surfaces. However, they can only produce static surfaces that are not controllable, which provides limited ability to modify the resulting model by editing its pose or shape parameters. Nevertheless, such features are essential in building flexible models for both computer graphics and computer vision. In this work, we present methodology that combines detail-rich implicit functions and parametric representations in order to reconstruct 3D models of people that remain controllable and accurate even in the presence of clothing. Given sparse 3D point clouds sampled on the surface of a dressed person, we use an Implicit Part Network (IP-Net)to jointly predict the outer 3D surface of the dressed person, the and inner body surface, and the semantic correspondences to a parametric body model. We subsequently use correspondences to fit the body model to our inner surface and then non-rigidly deform it (under a parametric body + displacement model) to the outer surface in order to capture garment, face and hair detail. In quantitative and qualitative experiments with both full body data and hand scans we show that the proposed methodology generalizes, and is effective even given incomplete point clouds collected from single-view depth images. Our models and code can be downloaded from http://virtualhumans.mpi-inf.mpg.de/ipnet.

Amir Rahimi, Thomas Mensink, Kartik Gupta et al.

Calibration of neural networks is a critical aspect to consider when incorporating machine learning models in real-world decision-making systems where the confidence of decisions are equally important as the decisions themselves. In recent years, there is a surge of research on neural network calibration and the majority of the works can be categorized into post-hoc calibration methods, defined as methods that learn an additional function to calibrate an already trained base network. In this work, we intend to understand the post-hoc calibration methods from a theoretical point of view. Especially, it is known that minimizing Negative Log-Likelihood (NLL) will lead to a calibrated network on the training set if the global optimum is attained (Bishop, 1994). Nevertheless, it is not clear learning an additional function in a post-hoc manner would lead to calibration in the theoretical sense. To this end, we prove that even though the base network ($f$) does not lead to the global optimum of NLL, by adding additional layers ($g$) and minimizing NLL by optimizing the parameters of $g$ one can obtain a calibrated network $g \circ f$. This not only provides a less stringent condition to obtain a calibrated network but also provides a theoretical justification of post-hoc calibration methods. Our experiments on various image classification benchmarks confirm the theory.

Yuhua Chen, Luc Van Gool, Cordelia Schmid et al.

Consistency Guided Scene Flow Estimation (CGSF) is a self-supervised framework for the joint reconstruction of 3D scene structure and motion from stereo video. The model takes two temporal stereo pairs as input, and predicts disparity and scene flow. The model self-adapts at test time by iteratively refining its predictions. The refinement process is guided by a consistency loss, which combines stereo and temporal photo-consistency with a geometric term that couples disparity and 3D motion. To handle inherent modeling error in the consistency loss (e.g. Lambertian assumptions) and for better generalization, we further introduce a learned, output refinement network, which takes the initial predictions, the loss, and the gradient as input, and efficiently predicts a correlated output update. In multiple experiments, including ablation studies, we show that the proposed model can reliably predict disparity and scene flow in challenging imagery, achieves better generalization than the state-of-the-art, and adapts quickly and robustly to unseen domains.

Alex Bewley, Pei Sun, Thomas Mensink et al.

This paper presents a novel 3D object detection framework that processes LiDAR data directly on its native representation: range images. Benefiting from the compactness of range images, 2D convolutions can efficiently process dense LiDAR data of a scene. To overcome scale sensitivity in this perspective view, a novel range-conditioned dilation (RCD) layer is proposed to dynamically adjust a continuous dilation rate as a function of the measured range. Furthermore, localized soft range gating combined with a 3D box-refinement stage improves robustness in occluded areas, and produces overall more accurate bounding box predictions. On the public large-scale Waymo Open Dataset, our method sets a new baseline for range-based 3D detection, outperforming multiview and voxel-based methods over all ranges with unparalleled performance at long range detection.

Andrei Zanfir, Eduard Gabriel Bazavan, Hongyi Xu et al.

Monocular 3D human pose and shape estimation is challenging due to the many degrees of freedom of the human body and thedifficulty to acquire training data for large-scale supervised learning in complex visual scenes. In this paper we present practical semi-supervised and self-supervised models that support training and good generalization in real-world images and video. Our formulation is based on kinematic latent normalizing flow representations and dynamics, as well as differentiable, semantic body part alignment loss functions that support self-supervised learning. In extensive experiments using 3D motion capture datasets like CMU, Human3.6M, 3DPW, or AMASS, as well as image repositories like COCO, we show that the proposed methods outperform the state of the art, supporting the practical construction of an accurate family of models based on large-scale training with diverse and incompletely labeled image and video data.

Erik Gärtner, Aleksis Pirinen, Cristian Sminchisescu

Most 3d human pose estimation methods assume that input -- be it images of a scene collected from one or several viewpoints, or from a video -- is given. Consequently, they focus on estimates leveraging prior knowledge and measurement by fusing information spatially and/or temporally, whenever available. In this paper we address the problem of an active observer with freedom to move and explore the scene spatially -- in `time-freeze' mode -- and/or temporally, by selecting informative viewpoints that improve its estimation accuracy. Towards this end, we introduce Pose-DRL, a fully trainable deep reinforcement learning-based active pose estimation architecture which learns to select appropriate views, in space and time, to feed an underlying monocular pose estimator. We evaluate our model using single- and multi-target estimators with strong result in both settings. Our system further learns automatic stopping conditions in time and transition functions to the next temporal processing step in videos. In extensive experiments with the Panoptic multi-view setup, and for complex scenes containing multiple people, we show that our model learns to select viewpoints that yield significantly more accurate pose estimates compared to strong multi-view baselines.

Henning Petzka, Michael Kamp, Linara Adilova et al.

Flatness of the loss curve is conjectured to be connected to the generalization ability of machine learning models, in particular neural networks. While it has been empirically observed that flatness measures consistently correlate strongly with generalization, it is still an open theoretical problem why and under which circumstances flatness is connected to generalization, in particular in light of reparameterizations that change certain flatness measures but leave generalization unchanged. We investigate the connection between flatness and generalization by relating it to the interpolation from representative data, deriving notions of representativeness, and feature robustness. The notions allow us to rigorously connect flatness and generalization and to identify conditions under which the connection holds. Moreover, they give rise to a novel, but natural relative flatness measure that correlates strongly with generalization, simplifies to ridge regression for ordinary least squares, and solves the reparameterization issue.

Henning Petzka, Linara Adilova, Michael Kamp et al.

The performance of deep neural networks is often attributed to their automated, task-related feature construction. It remains an open question, though, why this leads to solutions with good generalization, even in cases where the number of parameters is larger than the number of samples. Back in the 90s, Hochreiter and Schmidhuber observed that flatness of the loss surface around a local minimum correlates with low generalization error. For several flatness measures, this correlation has been empirically validated. However, it has recently been shown that existing measures of flatness cannot theoretically be related to generalization due to a lack of invariance with respect to reparameterizations. We propose a natural modification of existing flatness measures that results in invariance to reparameterization.

Mihai Zanfir, Elisabeta Oneata, Alin-Ionut Popa et al.

Generating good quality and geometrically plausible synthetic images of humans with the ability to control appearance, pose and shape parameters, has become increasingly important for a variety of tasks ranging from photo editing, fashion virtual try-on, to special effects and image compression. In this paper, we propose HUSC, a HUman Synthesis and Scene Compositing framework for the realistic synthesis of humans with different appearance, in novel poses and scenes. Central to our formulation is 3d reasoning for both people and scenes, in order to produce realistic collages, by correctly modeling perspective effects and occlusion, by taking into account scene semantics and by adequately handling relative scales. Conceptually our framework consists of three components: (1) a human image synthesis model with controllable pose and appearance, based on a parametric representation, (2) a person insertion procedure that leverages the geometry and semantics of the 3d scene, and (3) an appearance compositing process to create a seamless blending between the colors of the scene and the generated human image, and avoid visual artifacts. The performance of our framework is supported by both qualitative and quantitative results, in particular state-of-the art synthesis scores for the DeepFashion dataset.

Yuhua Chen, Cordelia Schmid, Cristian Sminchisescu

We present GLNet, a self-supervised framework for learning depth, optical flow, camera pose and intrinsic parameters from monocular video - addressing the difficulty of acquiring realistic ground-truth for such tasks. We propose three contributions: 1) we design new loss functions that capture multiple geometric constraints (eg. epipolar geometry) as well as an adaptive photometric loss that supports multiple moving objects, rigid and non-rigid, 2) we extend the model such that it predicts camera intrinsics, making it applicable to uncalibrated video, and 3) we propose several online refinement strategies that rely on the symmetry of our self-supervised loss in training and testing, in particular optimizing model parameters and/or the output of different tasks, thus leveraging their mutual interactions. The idea of jointly optimizing the system output, under all geometric and photometric constraints can be viewed as a dense generalization of classical bundle adjustment. We demonstrate the effectiveness of our method on KITTI and Cityscapes, where we outperform previous self-supervised approaches on multiple tasks. We also show good generalization for transfer learning in YouTube videos.

Henning Petzka, Cristian Sminchisescu

Understanding the loss surface of neural networks is essential for the design of models with predictable performance and their success in applications. Experimental results suggest that sufficiently deep and wide neural networks are not negatively impacted by suboptimal local minima. Despite recent progress, the reason for this outcome is not fully understood. Could deep networks have very few, if at all, suboptimal local optima? or could all of them be equally good? We provide a construction to show that suboptimal local minima (i.e., non-global ones), even though degenerate, exist for fully connected neural networks with sigmoid activation functions. The local minima obtained by our construction belong to a connected set of local solutions that can be escaped from via a non-increasing path on the loss curve. For extremely wide neural networks of decreasing width after the wide layer, we prove that every suboptimal local minimum belongs to such a connected set. This provides a partial explanation for the successful application of deep neural networks. In addition, we also characterize under what conditions the same construction leads to saddle points instead of local minima for deep neural networks.

Alin-Ionut Popa, Mihai Zanfir, Cristian Sminchisescu

We propose a deep multitask architecture for \emph{fully automatic 2d and 3d human sensing} (DMHS), including \emph{recognition and reconstruction}, in \emph{monocular images}. The system computes the figure-ground segmentation, semantically identifies the human body parts at pixel level, and estimates the 2d and 3d pose of the person. The model supports the joint training of all components by means of multi-task losses where early processing stages recursively feed into advanced ones for increasingly complex calculations, accuracy and robustness. The design allows us to tie a complete training protocol, by taking advantage of multiple datasets that would otherwise restrictively cover only some of the model components: complex 2d image data with no body part labeling and without associated 3d ground truth, or complex 3d data with limited 2d background variability. In detailed experiments based on several challenging 2d and 3d datasets (LSP, HumanEva, Human3.6M), we evaluate the sub-structures of the model, the effect of various types of training data in the multitask loss, and demonstrate that state-of-the-art results can be achieved at all processing levels. We also show that in the wild our monocular RGB architecture is perceptually competitive to a state-of-the art (commercial) Kinect system based on RGB-D data.

David Nilsson, Cristian Sminchisescu

Semantic video segmentation is challenging due to the sheer amount of data that needs to be processed and labeled in order to construct accurate models. In this paper we present a deep, end-to-end trainable methodology to video segmentation that is capable of leveraging information present in unlabeled data in order to improve semantic estimates. Our model combines a convolutional architecture and a spatio-temporal transformer recurrent layer that are able to temporally propagate labeling information by means of optical flow, adaptively gated based on its locally estimated uncertainty. The flow, the recognition and the gated temporal propagation modules can be trained jointly, end-to-end. The temporal, gated recurrent flow propagation component of our model can be plugged into any static semantic segmentation architecture and turn it into a weakly supervised video processing one. Our extensive experiments in the challenging CityScapes and Camvid datasets, and based on multiple deep architectures, indicate that the resulting model can leverage unlabeled temporal frames, next to a labeled one, in order to improve both the video segmentation accuracy and the consistency of its temporal labeling, at no additional annotation cost and with little extra computation.

Mihai Zanfir, Elisabeta Marinoiu, Cristian Sminchisescu

Automatic video captioning is challenging due to the complex interactions in dynamic real scenes. A comprehensive system would ultimately localize and track the objects, actions and interactions present in a video and generate a description that relies on temporal localization in order to ground the visual concepts. However, most existing automatic video captioning systems map from raw video data to high level textual description, bypassing localization and recognition, thus discarding potentially valuable information for content localization and generalization. In this work we present an automatic video captioning model that combines spatio-temporal attention and image classification by means of deep neural network structures based on long short-term memory. The resulting system is demonstrated to produce state-of-the-art results in the standard YouTube captioning benchmark while also offering the advantage of localizing the visual concepts (subjects, verbs, objects), with no grounding supervision, over space and time.