Fernando De la Torre

Chen Henry Wu, Fernando De la Torre · cmu

Diffusion models have achieved unprecedented performance in generative modeling. The commonly-adopted formulation of the latent code of diffusion models is a sequence of gradually denoised samples, as opposed to the simpler (e.g., Gaussian) latent space of GANs, VAEs, and normalizing flows. This paper provides an alternative, Gaussian formulation of the latent space of various diffusion models, as well as an invertible DPM-Encoder that maps images into the latent space. While our formulation is purely based on the definition of diffusion models, we demonstrate several intriguing consequences. (1) Empirically, we observe that a common latent space emerges from two diffusion models trained independently on related domains. In light of this finding, we propose CycleDiffusion, which uses DPM-Encoder for unpaired image-to-image translation. Furthermore, applying CycleDiffusion to text-to-image diffusion models, we show that large-scale text-to-image diffusion models can be used as zero-shot image-to-image editors. (2) One can guide pre-trained diffusion models and GANs by controlling the latent codes in a unified, plug-and-play formulation based on energy-based models. Using the CLIP model and a face recognition model as guidance, we demonstrate that diffusion models have better coverage of low-density sub-populations and individuals than GANs. The code is publicly available at https://github.com/ChenWu98/cycle-diffusion.

Chen Henry Wu, Saman Motamed, Shaunak Srivastava et al. · cmu

Generative models (e.g., GANs, diffusion models) learn the underlying data distribution in an unsupervised manner. However, many applications of interest require sampling from a particular region of the output space or sampling evenly over a range of characteristics. For efficient sampling in these scenarios, we propose Generative Visual Prompt (PromptGen), a framework for distributional control over pre-trained generative models by incorporating knowledge of other off-the-shelf models. PromptGen defines control as energy-based models (EBMs) and samples images in a feed-forward manner by approximating the EBM with invertible neural networks, avoiding optimization at inference. Our experiments demonstrate how PromptGen can efficiently sample from several unconditional generative models (e.g., StyleGAN2, StyleNeRF, diffusion autoencoder, NVAE) in a controlled or/and de-biased manner using various off-the-shelf models: (1) with the CLIP model as control, PromptGen can sample images guided by text, (2) with image classifiers as control, PromptGen can de-bias generative models across a set of attributes or attribute combinations, and (3) with inverse graphics models as control, PromptGen can sample images of the same identity in different poses. (4) Finally, PromptGen reveals that the CLIP model shows a "reporting bias" when used as control, and PromptGen can further de-bias this controlled distribution in an iterative manner. The code is available at https://github.com/ChenWu98/Generative-Visual-Prompt.

Cheng Zhang, Xuanbai Chen, Siqi Chai et al. · cmu

Text-to-image generative models often reflect the biases of the training data, leading to unequal representations of underrepresented groups. This study investigates inclusive text-to-image generative models that generate images based on human-written prompts and ensure the resulting images are uniformly distributed across attributes of interest. Unfortunately, directly expressing the desired attributes in the prompt often leads to sub-optimal results due to linguistic ambiguity or model misrepresentation. Hence, this paper proposes a drastically different approach that adheres to the maxim that "a picture is worth a thousand words". We show that, for some attributes, images can represent concepts more expressively than text. For instance, categories of skin tones are typically hard to specify by text but can be easily represented by example images. Building upon these insights, we propose a novel approach, ITI-GEN, that leverages readily available reference images for Inclusive Text-to-Image GENeration. The key idea is learning a set of prompt embeddings to generate images that can effectively represent all desired attribute categories. More importantly, ITI-GEN requires no model fine-tuning, making it computationally efficient to augment existing text-to-image models. Extensive experiments demonstrate that ITI-GEN largely improves over state-of-the-art models to generate inclusive images from a prompt. Project page: https://czhang0528.github.io/iti-gen.

Aashish Rai, Hiresh Gupta, Ayush Pandey et al.

In recent years, there has been significant progress in 2D generative face models fueled by applications such as animation, synthetic data generation, and digital avatars. However, due to the absence of 3D information, these 2D models often struggle to accurately disentangle facial attributes like pose, expression, and illumination, limiting their editing capabilities. To address this limitation, this paper proposes a 3D controllable generative face model to produce high-quality albedo and precise 3D shape leveraging existing 2D generative models. By combining 2D face generative models with semantic face manipulation, this method enables editing of detailed 3D rendered faces. The proposed framework utilizes an alternating descent optimization approach over shape and albedo. Differentiable rendering is used to train high-quality shapes and albedo without 3D supervision. Moreover, this approach outperforms the state-of-the-art (SOTA) methods in the well-known NoW benchmark for shape reconstruction. It also outperforms the SOTA reconstruction models in recovering rendered faces' identities across novel poses by an average of 10%. Additionally, the paper demonstrates direct control of expressions in 3D faces by exploiting latent space leading to text-based editing of 3D faces.

Fariborz Taherkhani, Aashish Rai, Quankai Gao et al.

3D face modeling has been an active area of research in computer vision and computer graphics, fueling applications ranging from facial expression transfer in virtual avatars to synthetic data generation. Existing 3D deep learning generative models (e.g., VAE, GANs) allow generating compact face representations (both shape and texture) that can model non-linearities in the shape and appearance space (e.g., scatter effects, specularities, etc.). However, they lack the capability to control the generation of subtle expressions. This paper proposes a new 3D face generative model that can decouple identity and expression and provides granular control over expressions. In particular, we propose using a pair of supervised auto-encoder and generative adversarial networks to produce high-quality 3D faces, both in terms of appearance and shape. Experimental results in the generation of 3D faces learned with holistic expression labels, or Action Unit labels, show how we can decouple identity and expression; gaining fine-control over expressions while preserving identity.

Yehonathan Litman, Or Patashnik, Kangle Deng et al.

Recent works in inverse rendering have shown promise in using multi-view images of an object to recover shape, albedo, and materials. However, the recovered components often fail to render accurately under new lighting conditions due to the intrinsic challenge of disentangling albedo and material properties from input images. To address this challenge, we introduce MaterialFusion, an enhanced conventional 3D inverse rendering pipeline that incorporates a 2D prior on texture and material properties. We present StableMaterial, a 2D diffusion model prior that refines multi-lit data to estimate the most likely albedo and material from given input appearances. This model is trained on albedo, material, and relit image data derived from a curated dataset of approximately ~12K artist-designed synthetic Blender objects called BlenderVault. we incorporate this diffusion prior with an inverse rendering framework where we use score distillation sampling (SDS) to guide the optimization of the albedo and materials, improving relighting performance in comparison with previous work. We validate MaterialFusion's relighting performance on 4 datasets of synthetic and real objects under diverse illumination conditions, showing our diffusion-aided approach significantly improves the appearance of reconstructed objects under novel lighting conditions. We intend to publicly release our BlenderVault dataset to support further research in this field.

Saman Motamed, Jianjin Xu, Chen Henry Wu et al.

Generative models such as StyleGAN2 and Stable Diffusion have achieved state-of-the-art performance in computer vision tasks such as image synthesis, inpainting, and de-noising. However, current generative models for face inpainting often fail to preserve fine facial details and the identity of the person, despite creating aesthetically convincing image structures and textures. In this work, we propose Person Aware Tuning (PAT) of Mask-Aware Transformer (MAT) for face inpainting, which addresses this issue. Our proposed method, PATMAT, effectively preserves identity by incorporating reference images of a subject and fine-tuning a MAT architecture trained on faces. By using ~40 reference images, PATMAT creates anchor points in MAT's style module, and tunes the model using the fixed anchors to adapt the model to a new face identity. Moreover, PATMAT's use of multiple images per anchor during training allows the model to use fewer reference images than competing methods. We demonstrate that PATMAT outperforms state-of-the-art models in terms of image quality, the preservation of person-specific details, and the identity of the subject. Our results suggest that PATMAT can be a promising approach for improving the quality of personalized face inpainting.

Youngjoong Kwon, Baole Fang, Yixing Lu et al.

Recent progress in neural rendering has brought forth pioneering methods, such as NeRF and Gaussian Splatting, which revolutionize view rendering across various domains like AR/VR, gaming, and content creation. While these methods excel at interpolating {\em within the training data}, the challenge of generalizing to new scenes and objects from very sparse views persists. Specifically, modeling 3D humans from sparse views presents formidable hurdles due to the inherent complexity of human geometry, resulting in inaccurate reconstructions of geometry and textures. To tackle this challenge, this paper leverages recent advancements in Gaussian Splatting and introduces a new method to learn generalizable human Gaussians that allows photorealistic and accurate view-rendering of a new human subject from a limited set of sparse views in a feed-forward manner. A pivotal innovation of our approach involves reformulating the learning of 3D Gaussian parameters into a regression process defined on the 2D UV space of a human template, which allows leveraging the strong geometry prior and the advantages of 2D convolutions. In addition, a multi-scaffold is proposed to effectively represent the offset details. Our method outperforms recent methods on both within-dataset generalization as well as cross-dataset generalization settings.

Haoye Dong, Aviral Chharia, Wenbo Gou et al.

3D Hand reconstruction from a single RGB image is challenging due to the articulated motion, self-occlusion, and interaction with objects. Existing SOTA methods employ attention-based transformers to learn the 3D hand pose and shape, yet they do not fully achieve robust and accurate performance, primarily due to inefficiently modeling spatial relations between joints. To address this problem, we propose a novel graph-guided Mamba framework, named Hamba, which bridges graph learning and state space modeling. Our core idea is to reformulate Mamba's scanning into graph-guided bidirectional scanning for 3D reconstruction using a few effective tokens. This enables us to efficiently learn the spatial relationships between joints for improving reconstruction performance. Specifically, we design a Graph-guided State Space (GSS) block that learns the graph-structured relations and spatial sequences of joints and uses 88.5% fewer tokens than attention-based methods. Additionally, we integrate the state space features and the global features using a fusion module. By utilizing the GSS block and the fusion module, Hamba effectively leverages the graph-guided state space features and jointly considers global and local features to improve performance. Experiments on several benchmarks and in-the-wild tests demonstrate that Hamba significantly outperforms existing SOTAs, achieving the PA-MPVPE of 5.3mm and F@15mm of 0.992 on FreiHAND. At the time of this paper's acceptance, Hamba holds the top position, Rank 1 in two Competition Leaderboards on 3D hand reconstruction. Project Website: https://humansensinglab.github.io/Hamba/

Gaoussou Youssouf Kebe, Jeffrey M. Girard, Einat Liebenthal et al.

This study introduces LlaMADRS, a novel framework leveraging open-source Large Language Models (LLMs) to automate depression severity assessment using the Montgomery-Asberg Depression Rating Scale (MADRS). We employ a zero-shot prompting strategy with carefully designed cues to guide the model in interpreting and scoring transcribed clinical interviews. Our approach, tested on 236 real-world interviews from the Context-Adaptive Multimodal Informatics (CAMI) dataset, demonstrates strong correlations with clinician assessments. The Qwen 2.5--72b model achieves near-human level agreement across most MADRS items, with Intraclass Correlation Coefficients (ICC) closely approaching those between human raters. We provide a comprehensive analysis of model performance across different MADRS items, highlighting strengths and current limitations. Our findings suggest that LLMs, with appropriate prompting, can serve as efficient tools for mental health assessment, potentially increasing accessibility in resource-limited settings. However, challenges remain, particularly in assessing symptoms that rely on non-verbal cues, underscoring the need for multimodal approaches in future work.

Rwiddhi Chakraborty, Yinong Wang, Jialu Gao et al.

The widespread success of deep learning models today is owed to the curation of extensive datasets significant in size and complexity. However, such models frequently pick up inherent biases in the data during the training process, leading to unreliable predictions. Diagnosing and debiasing datasets is thus a necessity to ensure reliable model performance. In this paper, we present ConBias, a novel framework for diagnosing and mitigating Concept co-occurrence Biases in visual datasets. ConBias represents visual datasets as knowledge graphs of concepts, enabling meticulous analysis of spurious concept co-occurrences to uncover concept imbalances across the whole dataset. Moreover, we show that by employing a novel clique-based concept balancing strategy, we can mitigate these imbalances, leading to enhanced performance on downstream tasks. Extensive experiments show that data augmentation based on a balanced concept distribution augmented by Conbias improves generalization performance across multiple datasets compared to state-of-the-art methods.

Prachi Garg, Joseph K J, Vineeth N Balasubramanian et al.

As extended reality (XR) is redefining how users interact with computing devices, research in human action recognition is gaining prominence. Typically, models deployed on immersive computing devices are static and limited to their default set of classes. The goal of our research is to provide users and developers with the capability to personalize their experience by adding new action classes to their device models continually. Importantly, a user should be able to add new classes in a low-shot and efficient manner, while this process should not require storing or replaying any of user's sensitive training data. We formalize this problem as privacy-aware few-shot continual action recognition. Towards this end, we propose POET: Prompt-Offset Tuning. While existing prompt tuning approaches have shown great promise for continual learning of image, text, and video modalities; they demand access to extensively pretrained transformers. Breaking away from this assumption, POET demonstrates the efficacy of prompt tuning a significantly lightweight backbone, pretrained exclusively on the base class data. We propose a novel spatio-temporal learnable prompt offset tuning approach, and are the first to apply such prompt tuning to Graph Neural Networks. We contribute two new benchmarks for our new problem setting in human action recognition: (i) NTU RGB+D dataset for activity recognition, and (ii) SHREC-2017 dataset for hand gesture recognition. We find that POET consistently outperforms comprehensive benchmarks. Source code at https://github.com/humansensinglab/POET-continual-action-recognition.

Alexander Richard, Michael Zollhoefer, Yandong Wen et al.

This paper presents a generic method for generating full facial 3D animation from speech. Existing approaches to audio-driven facial animation exhibit uncanny or static upper face animation, fail to produce accurate and plausible co-articulation or rely on person-specific models that limit their scalability. To improve upon existing models, we propose a generic audio-driven facial animation approach that achieves highly realistic motion synthesis results for the entire face. At the core of our approach is a categorical latent space for facial animation that disentangles audio-correlated and audio-uncorrelated information based on a novel cross-modality loss. Our approach ensures highly accurate lip motion, while also synthesizing plausible animation of the parts of the face that are uncorrelated to the audio signal, such as eye blinks and eye brow motion. We demonstrate that our approach outperforms several baselines and obtains state-of-the-art quality both qualitatively and quantitatively. A perceptual user study demonstrates that our approach is deemed more realistic than the current state-of-the-art in over 75% of cases. We recommend watching the supplemental video before reading the paper: https://github.com/facebookresearch/meshtalk

Jianjin Xu, Saman Motamed, Praneetha Vaddamanu et al.

Face inpainting is important in various applications, such as photo restoration, image editing, and virtual reality. Despite the significant advances in face generative models, ensuring that a person's unique facial identity is maintained during the inpainting process is still an elusive goal. Current state-of-the-art techniques, exemplified by MyStyle, necessitate resource-intensive fine-tuning and a substantial number of images for each new identity. Furthermore, existing methods often fall short in accommodating user-specified semantic attributes, such as beard or expression. To improve inpainting results, and reduce the computational complexity during inference, this paper proposes the use of Parallel Visual Attention (PVA) in conjunction with diffusion models. Specifically, we insert parallel attention matrices to each cross-attention module in the denoising network, which attends to features extracted from reference images by an identity encoder. We train the added attention modules and identity encoder on CelebAHQ-IDI, a dataset proposed for identity-preserving face inpainting. Experiments demonstrate that PVA attains unparalleled identity resemblance in both face inpainting and face inpainting with language guidance tasks, in comparison to various benchmarks, including MyStyle, Paint by Example, and Custom Diffusion. Our findings reveal that PVA ensures good identity preservation while offering effective language-controllability. Additionally, in contrast to Custom Diffusion, PVA requires just 40 fine-tuning steps for each new identity, which translates to a significant speed increase of over 20 times.

Or Patashnik, Rinon Gal, Daniel Cohen-Or et al.

Large-scale text-to-image models enable a wide range of image editing techniques, using text prompts or even spatial controls. However, applying these editing methods to multi-view images depicting a single scene leads to 3D-inconsistent results. In this work, we focus on spatial control-based geometric manipulations and introduce a method to consolidate the editing process across various views. We build on two insights: (1) maintaining consistent features throughout the generative process helps attain consistency in multi-view editing, and (2) the queries in self-attention layers significantly influence the image structure. Hence, we propose to improve the geometric consistency of the edited images by enforcing the consistency of the queries. To do so, we introduce QNeRF, a neural radiance field trained on the internal query features of the edited images. Once trained, QNeRF can render 3D-consistent queries, which are then softly injected back into the self-attention layers during generation, greatly improving multi-view consistency. We refine the process through a progressive, iterative method that better consolidates queries across the diffusion timesteps. We compare our method to a range of existing techniques and demonstrate that it can achieve better multi-view consistency and higher fidelity to the input scene. These advantages allow us to train NeRFs with fewer visual artifacts, that are better aligned with the target geometry.

Yixing Lu, Junting Dong, Youngjoong Kwon et al.

We present a unified and generalizable framework for synthesizing view-consistent and temporally coherent avatars from a single image, addressing the challenging task of single-image avatar generation. Existing diffusion-based methods often condition on sparse human templates (e.g., depth or normal maps), which leads to multi-view and temporal inconsistencies due to the mismatch between these signals and the true appearance of the subject. Our approach bridges this gap by combining the reconstruction power of regression-based 3D human reconstruction with the generative capabilities of a diffusion model. In a first step, an initial 3D reconstructed human through a generalized NeRF provides comprehensive conditioning, ensuring high-quality synthesis faithful to the reference appearance and structure. Subsequently, the derived geometry and appearance from the generalized NeRF serve as input to a video-based diffusion model. This strategic integration is pivotal for enforcing both multi-view and temporal consistency throughout the avatar's generation. Empirical results underscore the superior generalization ability of our proposed method, demonstrating its effectiveness across diverse in-domain and out-of-domain in-the-wild datasets.

Zoltán Á. Milacski, Koichiro Niinuma, Ryosuke Kawamura et al.

The connection between our 3D surroundings and the descriptive language that characterizes them would be well-suited for localizing and generating human motion in context but for one problem. The complexity introduced by multiple modalities makes capturing this connection challenging with a fixed set of descriptors. Specifically, closed vocabulary scene encoders, which require learning text-scene associations from scratch, have been favored in the literature, often resulting in inaccurate motion grounding. In this paper, we propose a method that integrates an open vocabulary scene encoder into the architecture, establishing a robust connection between text and scene. Our two-step approach starts with pretraining the scene encoder through knowledge distillation from an existing open vocabulary semantic image segmentation model, ensuring a shared text-scene feature space. Subsequently, the scene encoder is fine-tuned for conditional motion generation, incorporating two novel regularization losses that regress the category and size of the goal object. Our methodology achieves up to a 30% reduction in the goal object distance metric compared to the prior state-of-the-art baseline model on the HUMANISE dataset. This improvement is demonstrated through evaluations conducted using three implementations of our framework and a perceptual study. Additionally, our method is designed to seamlessly accommodate future 2D segmentation methods that provide per-pixel text-aligned features for distillation.

Chen Wu, Fernando De la Torre

Text-to-image diffusion models have achieved remarkable performance in image synthesis, while the text interface does not always provide fine-grained control over certain image factors. For instance, changing a single token in the text can have unintended effects on the image. This paper shows a simple modification of classifier-free guidance can help disentangle image factors in text-to-image models. The key idea of our method, Contrastive Guidance, is to characterize an intended factor with two prompts that differ in minimal tokens: the positive prompt describes the image to be synthesized, and the baseline prompt serves as a "baseline" that disentangles other factors. Contrastive Guidance is a general method we illustrate whose benefits in three scenarios: (1) to guide domain-specific diffusion models trained on an object class, (2) to gain continuous, rig-like controls for text-to-image generation, and (3) to improve the performance of zero-shot image editors.

Yehonathan Litman, Fernando De la Torre, Shubham Tulsiani

Recent approaches for 3D relighting have shown promise in integrating 2D image relighting generative priors to alter the appearance of a 3D representation while preserving the underlying structure. Nevertheless, generative priors used for 2D relighting that directly relight from an input image do not take advantage of intrinsic properties of the subject that can be inferred or cannot consider multi-view data at scale, leading to subpar relighting. In this paper, we propose Lightswitch, a novel finetuned material-relighting diffusion framework that efficiently relights an arbitrary number of input images to a target lighting condition while incorporating cues from inferred intrinsic properties. By using multi-view and material information cues together with a scalable denoising scheme, our method consistently and efficiently relights dense multi-view data of objects with diverse material compositions. We show that our 2D relighting prediction quality exceeds previous state-of-the-art relighting priors that directly relight from images. We further demonstrate that LightSwitch matches or outperforms state-of-the-art diffusion inverse rendering methods in relighting synthetic and real objects in as little as 2 minutes.

Runkai Zheng, Vishnu Asutosh Dasu, Yinong Oliver Wang et al.

Modern machine learning models heavily rely on large datasets that often include sensitive and private information, raising serious privacy concerns. Differentially private (DP) data generation offers a solution by creating synthetic datasets that limit the leakage of private information within a predefined privacy budget; however, it requires a substantial amount of data to achieve performance comparable to models trained on the original data. To mitigate the significant expense incurred with synthetic data generation, Dataset Distillation (DD) stands out for its remarkable training and storage efficiency. This efficiency is particularly advantageous when integrated with DP mechanisms, curating compact yet informative synthetic datasets without compromising privacy. However, current state-of-the-art private DD methods suffer from a synchronized sampling-optimization process and the dependency on noisy training signals from randomly initialized networks. This results in the inefficient utilization of private information due to the addition of excessive noise. To address these issues, we introduce a novel framework that decouples sampling from optimization for better convergence and improves signal quality by mitigating the impact of DP noise through matching in an informative subspace. On CIFAR-10, our method achieves a \textbf{10.0\%} improvement with 50 images per class and \textbf{8.3\%} increase with just \textbf{one-fifth} the distilled set size of previous state-of-the-art methods, demonstrating significant potential to advance privacy-preserving DD.

Saswat Subhajyoti Mallick, Rahul Goel, Bernhard Kerbl et al.

3D Gaussian Splatting (3DGS) has transformed novel-view synthesis with its fast, interpretable, and high-fidelity rendering. However, its resource requirements limit its usability. Especially on constrained devices, training performance degrades quickly and often cannot complete due to excessive memory consumption of the model. The method converges with an indefinite number of Gaussians -- many of them redundant -- making rendering unnecessarily slow and preventing its usage in downstream tasks that expect fixed-size inputs. To address these issues, we tackle the challenges of training and rendering 3DGS models on a budget. We use a guided, purely constructive densification process that steers densification toward Gaussians that raise the reconstruction quality. Model size continuously increases in a controlled manner towards an exact budget, using score-based densification of Gaussians with training-time priors that measure their contribution. We further address training speed obstacles: following a careful analysis of 3DGS' original pipeline, we derive faster, numerically equivalent solutions for gradient computation and attribute updates, including an alternative parallelization for efficient backpropagation. We also propose quality-preserving approximations where suitable to reduce training time even further. Taken together, these enhancements yield a robust, scalable solution with reduced training times, lower compute and memory requirements, and high quality. Our evaluation shows that in a budgeted setting, we obtain competitive quality metrics with 3DGS while achieving a 4--5x reduction in both model size and training time. With more generous budgets, our measured quality surpasses theirs. These advances open the door for novel-view synthesis in constrained environments, e.g., mobile devices.

Amin Jourabloo, Baris Gecer, Fernando De la Torre et al.

Social presence, the feeling of being there with a real person, will fuel the next generation of communication systems driven by digital humans in virtual reality (VR). The best 3D video-realistic VR avatars that minimize the uncanny effect rely on person-specific (PS) models. However, these PS models are time-consuming to build and are typically trained with limited data variability, which results in poor generalization and robustness. Major sources of variability that affects the accuracy of facial expression transfer algorithms include using different VR headsets (e.g., camera configuration, slop of the headset), facial appearance changes over time (e.g., beard, make-up), and environmental factors (e.g., lighting, backgrounds). This is a major drawback for the scalability of these models in VR. This paper makes progress in overcoming these limitations by proposing an end-to-end multi-identity architecture (MIA) trained with specialized augmentation strategies. MIA drives the shape component of the avatar from three cameras in the VR headset (two eyes, one mouth), in untrained subjects, using minimal personalized information (i.e., neutral 3D mesh shape). Similarly, if the PS texture decoder is available, MIA is able to drive the full avatar (shape+texture) robustly outperforming PS models in challenging scenarios. Our key contribution to improve robustness and generalization, is that our method implicitly decouples, in an unsupervised manner, the facial expression from nuisance factors (e.g., headset, environment, facial appearance). We demonstrate the superior performance and robustness of the proposed method versus state-of-the-art PS approaches in a variety of experiments.

Shugao Ma, Tomas Simon, Jason Saragih et al.

Telecommunication with photorealistic avatars in virtual or augmented reality is a promising path for achieving authentic face-to-face communication in 3D over remote physical distances. In this work, we present the Pixel Codec Avatars (PiCA): a deep generative model of 3D human faces that achieves state of the art reconstruction performance while being computationally efficient and adaptive to the rendering conditions during execution. Our model combines two core ideas: (1) a fully convolutional architecture for decoding spatially varying features, and (2) a rendering-adaptive per-pixel decoder. Both techniques are integrated via a dense surface representation that is learned in a weakly-supervised manner from low-topology mesh tracking over training images. We demonstrate that PiCA improves reconstruction over existing techniques across testing expressions and views on persons of different gender and skin tone. Importantly, we show that the PiCA model is much smaller than the state-of-art baseline model, and makes multi-person telecommunicaiton possible: on a single Oculus Quest 2 mobile VR headset, 5 avatars are rendered in realtime in the same scene.

Lele Chen, Chen Cao, Fernando De la Torre et al.

3D video avatars can empower virtual communications by providing compression, privacy, entertainment, and a sense of presence in AR/VR. Best 3D photo-realistic AR/VR avatars driven by video, that can minimize uncanny effects, rely on person-specific models. However, existing person-specific photo-realistic 3D models are not robust to lighting, hence their results typically miss subtle facial behaviors and cause artifacts in the avatar. This is a major drawback for the scalability of these models in communication systems (e.g., Messenger, Skype, FaceTime) and AR/VR. This paper addresses previous limitations by learning a deep learning lighting model, that in combination with a high-quality 3D face tracking algorithm, provides a method for subtle and robust facial motion transfer from a regular video to a 3D photo-realistic avatar. Extensive experimental validation and comparisons to other state-of-the-art methods demonstrate the effectiveness of the proposed framework in real-world scenarios with variability in pose, expression, and illumination. Please visit https://www.youtube.com/watch?v=dtz1LgZR8cc for more results. Our project page can be found at https://www.cs.rochester.edu/u/lchen63.

Xiangyu Xu, Hao Chen, Francesc Moreno-Noguer et al.

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.

Matthew Dale, David Griffin, Richard F. L. Evans et al.

Advances in artificial intelligence are driven by technologies inspired by the brain, but these technologies are orders of magnitude less powerful and energy efficient than biological systems. Inspired by the nonlinear dynamics of neural networks, new unconventional computing hardware has emerged with the potential to exploit natural phenomena and gain efficiency, in a similar manner to biological systems. Physical reservoir computing demonstrates this with a variety of unconventional systems, from optical-based to memristive systems. Reservoir computers provide a nonlinear projection of the task input into a high-dimensional feature space by exploiting the system's internal dynamics. A trained readout layer then combines features to perform tasks, such as pattern recognition and time-series analysis. Despite progress, achieving state-of-the-art performance without external signal processing to the reservoir remains challenging. Here we perform an initial exploration of three magnetic materials in thin-film geometries via microscale simulation. Our results reveal that basic spin properties of magnetic films generate the required nonlinear dynamics and memory to solve machine learning tasks (although there would be practical challenges in exploiting these particular materials in physical implementations). The method of exploration can be applied to other materials, so this work opens up the possibility of testing different materials, from relatively simple (alloys) to significantly complex (antiferromagnetic reservoirs).

Denis Tome, Thiemo Alldieck, Patrick Peluse et al.

We present a solution to egocentric 3D body pose estimation from monocular images captured from downward looking fish-eye cameras installed on the rim of a head mounted VR device. This unusual viewpoint leads to images with unique visual appearance, with severe self-occlusions and perspective distortions that result in drastic differences in resolution between lower and upper body. We propose an encoder-decoder architecture with a novel multi-branch decoder designed to account for the varying uncertainty in 2D predictions. The quantitative evaluation, on synthetic and real-world datasets, shows that our strategy leads to substantial improvements in accuracy over state of the art egocentric approaches. To tackle the lack of labelled data we also introduced a large photo-realistic synthetic dataset. xR-EgoPose offers high quality renderings of people with diverse skintones, body shapes and clothing, performing a range of actions. Our experiments show that the high variability in our new synthetic training corpus leads to good generalization to real world footage and to state of theart results on real world datasets with ground truth. Moreover, an evaluation on the Human3.6M benchmark shows that the performance of our method is on par with top performing approaches on the more classic problem of 3D human pose from a third person viewpoint.

Hang Chu, Shugao Ma, Fernando De la Torre et al.

VR telepresence consists of interacting with another human in a virtual space represented by an avatar. Today most avatars are cartoon-like, but soon the technology will allow video-realistic ones. This paper aims in this direction and presents Modular Codec Avatars (MCA), a method to generate hyper-realistic faces driven by the cameras in the VR headset. MCA extends traditional Codec Avatars (CA) by replacing the holistic models with a learned modular representation. It is important to note that traditional person-specific CAs are learned from few training samples, and typically lack robustness as well as limited expressiveness when transferring facial expressions. MCAs solve these issues by learning a modulated adaptive blending of different facial components as well as an exemplar-based latent alignment. We demonstrate that MCA achieves improved expressiveness and robustness w.r.t to CA in a variety of real-world datasets and practical scenarios. Finally, we showcase new applications in VR telepresence enabled by the proposed model.

Alexander Richard, Colin Lea, Shugao Ma et al.

Codec Avatars are a recent class of learned, photorealistic face models that accurately represent the geometry and texture of a person in 3D (i.e., for virtual reality), and are almost indistinguishable from video. In this paper we describe the first approach to animate these parametric models in real-time which could be deployed on commodity virtual reality hardware using audio and/or eye tracking. Our goal is to display expressive conversations between individuals that exhibit important social signals such as laughter and excitement solely from latent cues in our lossy input signals. To this end we collected over 5 hours of high frame rate 3D face scans across three participants including traditional neutral speech as well as expressive and conversational speech. We investigate a multimodal fusion approach that dynamically identifies which sensor encoding should animate which parts of the face at any time. See the supplemental video which demonstrates our ability to generate full face motion far beyond the typically neutral lip articulations seen in competing work: https://research.fb.com/videos/audio-and-gaze-driven-facial-animation-of-codec-avatars/

Xiangyu Xu, Hao Chen, Francesc Moreno-Noguer et al.

3D human shape and pose estimation from monocular images has been an active area of research in computer vision, having a substantial impact on the development of new applications, from activity recognition to creating virtual avatars. Existing deep learning methods for 3D human shape and pose estimation rely on relatively high-resolution input images; however, high-resolution visual content is not always available in several practical scenarios such as video surveillance and sports broadcasting. Low-resolution images in real scenarios can vary in a wide range of sizes, and a model trained in one resolution does not typically degrade gracefully across resolutions. Two common approaches to solve the problem of low-resolution input are applying super-resolution techniques to the input images which may result in visual 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 network is able to learn the 3D body shape and pose across different resolutions with a single model. The self-supervision loss encourages scale-consistency of the output, and the contrastive learning scheme enforces scale-consistency of the deep features. We show that both these new training losses provide robustness when learning 3D shape and pose in a weakly-supervised manner. Extensive experiments demonstrate that the RSC-Net can achieve consistently better results than the state-of-the-art methods for challenging low-resolution images.

Stanislav Panev, Francisco Vicente, Fernando De la Torre et al.

We propose a robust method for estimating road curb 3D parameters (size, location, orientation) using a calibrated monocular camera equipped with a fisheye lens. Automatic curb detection and localization is particularly important in the context of Advanced Driver Assistance System (ADAS), i.e. to prevent possible collision and damage of the vehicle's bumper during perpendicular and diagonal parking maneuvers. Combining 3D geometric reasoning with advanced vision-based detection methods, our approach is able to estimate the vehicle to curb distance in real time with mean accuracy of more than 90%, as well as its orientation, height and depth. Our approach consists of two distinct components - curb detection in each individual video frame and temporal analysis. The first part comprises of sophisticated curb edges extraction and parametrized 3D curb template fitting. Using a few assumptions regarding the real world geometry, we can thus retrieve the curb's height and its relative position w.r.t. the moving vehicle on which the camera is mounted. Support Vector Machine (SVM) classifier fed with Histograms of Oriented Gradients (HOG) is used for appearance-based filtering out outliers. In the second part, the detected curb regions are tracked in the temporal domain, so as to perform a second pass of false positives rejection. We have validated our approach on a newly collected database of 11 videos under different conditions. We have used point-wise LIDAR measurements and manual exhaustive labels as a ground truth.

J. D. Curtó, I. C. Zarza, Fernando de la Torre et al.

Generative Adversarial Networks (GANs) [Goodfellow et al. 2014] convergence in a high-resolution setting with a computational constrain of GPU memory capacity has been beset with difficulty due to the known lack of convergence rate stability. In order to boost network convergence of DCGAN (Deep Convolutional Generative Adversarial Networks) [Radford et al. 2016] and achieve good-looking high-resolution results we propose a new layered network, HDCGAN, that incorporates current state-of-the-art techniques for this effect. Glasses, a mechanism to arbitrarily improve the final GAN generated results by enlarging the input size by a telescope ζ is also presented. A novel bias-free dataset, Curtó & Zarza, containing human faces from different ethnical groups in a wide variety of illumination conditions and image resolutions is introduced. Curtó is enhanced with HDCGAN synthetic images, thus being the first GAN augmented dataset of faces. We conduct extensive experiments on CelebA [Liu et al. 2015], CelebA-hq [Karras et al. 2018] and Curtó. HDCGAN is the current state-of-the-art in synthetic image generation on CelebA achieving a MS-SSIM of 0.1978 and a FRÉCHET Inception Distance of 8.44.

J. D. Curtó, I. C. Zarza, Feng Yang et al.

McKernel introduces a framework to use kernel approximates in the mini-batch setting with Stochastic Gradient Descent (SGD) as an alternative to Deep Learning. Based on Random Kitchen Sinks [Rahimi and Recht 2007], we provide a C++ library for Large-scale Machine Learning. It contains a CPU optimized implementation of the algorithm in [Le et al. 2013], that allows the computation of approximated kernel expansions in log-linear time. The algorithm requires to compute the product of matrices Walsh Hadamard. A cache friendly Fast Walsh Hadamard that achieves compelling speed and outperforms current state-of-the-art methods has been developed. McKernel establishes the foundation of a new architecture of learning that allows to obtain large-scale non-linear classification combining lightning kernel expansions and a linear classifier. It travails in the mini-batch setting working analogously to Neural Networks. We show the validity of our method through extensive experiments on MNIST and FASHION MNIST [Xiao et al. 2017].

Enrique Sánchez-Lozano, Georgios Tzimiropoulos, Brais Martinez et al.

Linear regression is a fundamental building block in many face detection and tracking algorithms, typically used to predict shape displacements from image features through a linear mapping. This paper presents a Functional Regression solution to the least squares problem, which we coin Continuous Regression, resulting in the first real-time incremental face tracker. Contrary to prior work in Functional Regression, in which B-splines or Fourier series were used, we propose to approximate the input space by its first-order Taylor expansion, yielding a closed-form solution for the continuous domain of displacements. We then extend the continuous least squares problem to correlated variables, and demonstrate the generalisation of our approach. We incorporate Continuous Regression into the cascaded regression framework, and show its computational benefits for both training and testing. We then present a fast approach for incremental learning within Cascaded Continuous Regression, coined iCCR, and show that its complexity allows real-time face tracking, being 20 times faster than the state of the art. To the best of our knowledge, this is the first incremental face tracker that is shown to operate in real-time. We show that iCCR achieves state-of-the-art performance on the 300-VW dataset, the most recent, large-scale benchmark for face tracking.

Ramon Sanabria, Florian Metze, Fernando De La Torre

Speech is one of the most effective ways of communication among humans. Even though audio is the most common way of transmitting speech, very important information can be found in other modalities, such as vision. Vision is particularly useful when the acoustic signal is corrupted. Multi-modal speech recognition however has not yet found wide-spread use, mostly because the temporal alignment and fusion of the different information sources is challenging. This paper presents an end-to-end audiovisual speech recognizer (AVSR), based on recurrent neural networks (RNN) with a connectionist temporal classification (CTC) loss function. CTC creates sparse "peaky" output activations, and we analyze the differences in the alignments of output targets (phonemes or visemes) between audio-only, video-only, and audio-visual feature representations. We present the first such experiments on the large vocabulary IBM ViaVoice database, which outperform previously published approaches on phone accuracy in clean and noisy conditions.

Ji Zhao, Liantao Wang, Ricardo Cabral et al.

Visual learning problems such as object classification and action recognition are typically approached using extensions of the popular bag-of-words (BoW) model. Despite its great success, it is unclear what visual features the BoW model is learning: Which regions in the image or video are used to discriminate among classes? Which are the most discriminative visual words? Answering these questions is fundamental for understanding existing BoW models and inspiring better models for visual recognition. To answer these questions, this paper presents a method for feature selection and region selection in the visual BoW model. This allows for an intermediate visualization of the features and regions that are important for visual learning. The main idea is to assign latent weights to the features or regions, and jointly optimize these latent variables with the parameters of a classifier (e.g., support vector machine). There are four main benefits of our approach: (1) Our approach accommodates non-linear additive kernels such as the popular $χ^2$ and intersection kernel; (2) our approach is able to handle both regions in images and spatio-temporal regions in videos in a unified way; (3) the feature selection problem is convex, and both problems can be solved using a scalable reduced gradient method; (4) we point out strong connections with multiple kernel learning and multiple instance learning approaches. Experimental results in the PASCAL VOC 2007, MSR Action Dataset II and YouTube illustrate the benefits of our approach.