Reconstructing real-world 3D objects has numerous applications in computer vision, such as virtual reality, video games, and animations. Ideally, 3D reconstruction methods should generate high-fidelity results with 3D consistency in real-time. Traditional methods match pixels between images using photo-consistency constraints or learned features, while differentiable rendering methods like Neural Radiance Fields (NeRF) use differentiable volume rendering or surface-based representation to generate high-fidelity scenes. However, these methods require excessive runtime for rendering, making them impractical for daily applications. To address these challenges, we present EvaSurf, an Efficient View-Aware implicit textured Surface reconstruction method on mobile devices. In our method, we first employ an efficient surface-based model with a multi-view supervision module to ensure accurate mesh reconstruction. To enable high-fidelity rendering, we learn an implicit texture embedded with view-aware encoding to capture view-dependent information. Furthermore, with the explicit geometry and the implicit texture, we can employ a lightweight neural shader to reduce the expense of computation and further support real-time rendering on common mobile devices. Extensive experiments demonstrate that our method can reconstruct high-quality appearance and accurate mesh on both synthetic and real-world datasets. Moreover, our method can be trained in just 1-2 hours using a single GPU and run on mobile devices at over 40 FPS (Frames Per Second), with a final package required for rendering taking up only 40-50 MB.

Reconstructing objects from posed images is a crucial and complex task in computer graphics and computer vision. While NeRF-based neural reconstruction methods have exhibited impressive reconstruction ability, they tend to be time-comsuming. Recent strategies have adopted 3D Gaussian Splatting (3D-GS) for inverse rendering, which have led to quick and effective outcomes. However, these techniques generally have difficulty in producing believable geometries and materials for glossy objects, a challenge that stems from the inherent ambiguities of inverse rendering. To address this, we introduce GlossyGS, an innovative 3D-GS-based inverse rendering framework that aims to precisely reconstruct the geometry and materials of glossy objects by integrating material priors. The key idea is the use of micro-facet geometry segmentation prior, which helps to reduce the intrinsic ambiguities and improve the decomposition of geometries and materials. Additionally, we introduce a normal map prefiltering strategy to more accurately simulate the normal distribution of reflective surfaces. These strategies are integrated into a hybrid geometry and material representation that employs both explicit and implicit methods to depict glossy objects. We demonstrate through quantitative analysis and qualitative visualization that the proposed method is effective to reconstruct high-fidelity geometries and materials of glossy objects, and performs favorably against State-of-the-Arts.

Quantizing deep networks with adaptive bit-widths is a promising technique for efficient inference across many devices and resource constraints. In contrast to static methods that repeat the quantization process and train different models for different constraints, adaptive quantization enables us to flexibly adjust the bit-widths of a single deep network during inference for instant adaptation in different scenarios. While existing research shows encouraging results on common image classification benchmarks, this paper investigates how to train such adaptive networks more effectively. Specifically, we present two novel techniques for quantizing deep neural networks with adaptive bit-widths of weights and activations. First, we propose a collaborative strategy to choose a high-precision "teacher" for transferring knowledge to the low-precision "student" while jointly optimizing the model with all bit-widths. Second, to effectively transfer knowledge, we develop a dynamic block swapping method by randomly replacing the blocks in the lower-precision student network with the corresponding blocks in the higher-precision teacher network. Extensive experiments on multiple image and video classification datasets, well demonstrate the efficacy of our approach over state-of-the-art methods.

Head generation with diverse identities is an important task in computer vision and computer graphics, widely used in multimedia applications. However, current full-head generation methods require a large number of three-dimensional (3D) scans or multi-view images to train the model, resulting in expensive data acquisition costs. To address this issue, we propose Head3D, a method to generate full 3D heads with limited multi-view images. Specifically, our approach first extracts facial priors represented by tri-planes learned in EG3D, a 3D-aware generative model, and then proposes feature distillation to deliver the 3D frontal faces within complete heads without compromising head integrity. To mitigate the domain gap between the face and head models, we present a dual-discriminator to guide the frontal and back head generation. Our model achieves cost-efficient and diverse complete head generation with photo-realistic renderings and high-quality geometry representations. Extensive experiments demonstrate the effectiveness of our proposed Head3D, both qualitatively and quantitatively.

Building an interactive AI assistant that can perceive, reason, and collaborate with humans in the real world has been a long-standing pursuit in the AI community. This work is part of a broader research effort to develop intelligent agents that can interactively guide humans through performing tasks in the physical world. As a first step in this direction, we introduce HoloAssist, a large-scale egocentric human interaction dataset, where two people collaboratively complete physical manipulation tasks. The task performer executes the task while wearing a mixed-reality headset that captures seven synchronized data streams. The task instructor watches the performer’s egocentric video in real time and guides them verbally. By augmenting the data with action and conversational annotations and observing the rich behaviors of various participants, we present key insights into how human assistants correct mistakes, intervene in the task completion procedure, and ground their instructions to the environment. HoloAssist spans 166 hours of data captured by 350 unique instructor-performer pairs. Furthermore, we construct and present benchmarks on mistake detection, intervention type prediction, and hand forecasting, along with detailed analysis. We expect HoloAssist will provide an important resource for building AI assistants that can fluidly collaborate with humans in the real world. Data can be downloaded at https://holoassist.github.io/.

In 3D face reconstruction, orthogonal projection has been widely employed to substitute perspective projection to simplify the fitting process. This approximation performs well when the distance between camera and face is far enough. However, in some scenarios that the face is very close to camera or moving along the camera axis, the methods suffer from the inaccurate reconstruction and unstable temporal fitting due to the distortion under the perspective projection. In this paper, we aim to address the problem of single-image 3D face reconstruction under perspective projection. Specifically, a deep neural network, Perspective Network (PerspNet), is proposed to simultaneously reconstruct 3D face shape in canonical space and learn the correspondence between 2D pixels and 3D points, by which the 6DoF (6 Degrees of Freedom) face pose can be estimated to represent perspective projection. Besides, we contribute a large ARKitFace dataset to enable the training and evaluation of 3D face reconstruction solutions under the scenarios of perspective projection, which has 902,724 2D facial images with ground-truth 3D face mesh and annotated 6DoF pose parameters. Experimental results show that our approach outperforms current state-of-the-art methods by a significant margin. The code and data are available at https://github.com/cbsropenproject/6dof_face.

While implicit representations have achieved high-fidelity results in 3D rendering, it remains challenging to deforming and animating the implicit field. Existing works typically leverage data-dependent models as deformation priors, such as SMPL for human body animation. However, this dependency on category-specific priors limits them to generalize to other objects. To solve this problem, we propose a novel framework for deforming and animating the neural radiance field learned on \textit{arbitrary} objects. The key insight is that we introduce a cage-based representation as deformation prior, which is category-agnostic. Specifically, the deformation is performed based on an enclosing polygon mesh with sparsely defined vertices called \textit{cage} inside the rendering space, where each point is projected into a novel position based on the barycentric interpolation of the deformed cage vertices. In this way, we transform the cage into a generalized constraint, which is able to deform and animate arbitrary target objects while preserving geometry details. Based on extensive experiments, we demonstrate the effectiveness of our framework in the task of geometry editing, object animation and deformation transfer.

Videos capture events that typically contain multiple sequential, and simultaneous, actions even in the span of only a few seconds. However, most large-scale datasets built to train models for action recognition in video only provide a single label per video. Consequently, models can be incorrectly penalized for classifying actions that exist in the videos but are not explicitly labeled and do not learn the full spectrum of information present in each video in training. Towards this goal, we present the Multi-Moments in Time dataset (M-MiT) which includes over two million action labels for over one million three second videos. This multi-label dataset introduces novel challenges on how to train and analyze models for multi-action detection. Here, we present baseline results for multi-action recognition using loss functions adapted for long tail multi-label learning, provide improved methods for visualizing and interpreting models trained for multi-label action detection and show the strength of transferring models trained on M-MiT to smaller datasets.

The self-attention-based model, transformer, is recently becoming the leading backbone in the field of computer vision. In spite of the impressive success made by transformers in a variety of vision tasks, it still suffers from heavy computation and intensive memory costs. To address this limitation, this paper presents an Interpretability-Aware REDundancy REDuction framework (IA-RED$^2$). We start by observing a large amount of redundant computation, mainly spent on uncorrelated input patches, and then introduce an interpretable module to dynamically and gracefully drop these redundant patches. This novel framework is then extended to a hierarchical structure, where uncorrelated tokens at different stages are gradually removed, resulting in a considerable shrinkage of computational cost. We include extensive experiments on both image and video tasks, where our method could deliver up to 1.4x speed-up for state-of-the-art models like DeiT and TimeSformer, by only sacrificing less than 0.7% accuracy. More importantly, contrary to other acceleration approaches, our method is inherently interpretable with substantial visual evidence, making vision transformer closer to a more human-understandable architecture while being lighter. We demonstrate that the interpretability that naturally emerged in our framework can outperform the raw attention learned by the original visual transformer, as well as those generated by off-the-shelf interpretation methods, with both qualitative and quantitative results. Project Page: http://people.csail.mit.edu/bpan/ia-red/.

Research on facial action unit (AU) analysis typically require facial images that are labelled with those action units. While unlabelled facial images abound, labelling those images with action units or intensity is costly and time-consuming. Our approach makes it possible to analyze facial AUs when only some of the images have been labelled. We use many facial images to learn a deep framework that is able to take advantage of the facial representations. A restricted Boltzmann machine uses the available AU annotations to learn the AU label or intensity distribution. We train a support vector machine for AU recognition and a support vector regression for AU intensity estimation by maximizing the log likelihood of the AU mapping functions, taking into account the learned multiple AU distribution for all training data, while simultaneously diminishing errors between the predicted action units and ground-truth action unit occurrence or intensities for all labelled data. We perform experiments on two databases. The results demonstrate the superiority of a deep neural network for learning facial features, as well as the benefit of action unit label or intensity constraints for action unit occurrence recognition or intensity estimation in fully or semi-supervised scenarios.

The inherent connections among aesthetic attributes and aesthetics are crucial for image aesthetic assessment, but have not been thoroughly explored yet. In this paper, we propose a novel image aesthetic assessment assisted by attributes through both representation-level and label-level. The attributes are used as privileged information, which is only required during training. Specifically, we first propose a multitask deep convolutional rating network to learn the aesthetic score and attributes simultaneously. The attributes are explored to construct better feature representations for aesthetic assessment through multi-task learning. After that, we introduce a discriminator to distinguish the predicted attributes and aesthetics of the multi-task deep network from the ground truth label distribution embedded in the training data. The multi-task deep network wants to output aesthetic score and attributes as close to the ground truth labels as possible. Thus the deep network and the discriminator compete with each other. Through adversarial learning, the attributes are explored to enforce the distribution of the predicted attributes and aesthetics to converge to the ground truth label distribution. Experimental results on two benchmark databases demonstrate the superiority of the proposed method to state of the art work.

The spatial and temporal patterns inherent in facial feature points are crucial for facial landmark tracking, but have not been thoroughly explored yet. In this paper, we propose a novel deep adversarial framework to explore the shape and temporal dependencies from both appearance level and target label level. The proposed deep adversarial framework consists of a deep landmark tracker and a discriminator. The deep landmark tracker is composed of a stacked Hourglass network as well as a convolutional neural network and a long short-term memory network, and thus implicitly capture spatial and temporal patterns from facial appearance for facial landmark tracking. The discriminator is adopted to distinguish the tracked facial landmarks from ground truth ones. It explicitly models shape and temporal dependencies existing in ground truth facial landmarks through another convolutional neural network and another long short-term memory network. The deep landmark tracker and the discriminator compete with each other. Through adversarial learning, the proposed deep adversarial landmark tracking approach leverages inherent spatial and temporal patterns to facilitate facial landmark tracking from both appearance level and target label level. Experimental results on two benchmark databases demonstrate the superiority of the proposed approach to state-of-the-art work.

In this paper, we propose a novel approach of occluded facial expression recognition under the help of non-occluded facial images. The non-occluded facial images are used as privileged information, which is only required during training, but not required during testing. Specifically, two deep neural networks are first trained from occluded and non-occluded facial images respectively. Then the non-occluded network is fixed and is used to guide the fine-tuning of the occluded network from both label space and feature space. Similarity constraint and loss inequality regularization are imposed to the label space to make the output of occluded network converge to that of the non-occluded network. Adversarial leaning is adopted to force the distribution of the learned features from occluded facial images to be close to that from non-occluded facial images. Furthermore, a decoder network is employed to reconstruct the non-occluded facial images from occluded features. Under the guidance of non-occluded facial images, the occluded network is expected to learn better features and classifier during training. Experiments on the benchmark databases with both synthesized and realistic occluded facial images demonstrate the superiority of the proposed method to state-of-the-art.

Deep convolutional neural networks (CNNs) have made impressive progress in many video recognition tasks such as video pose estimation and video object detection. However, CNN inference on video is computationally expensive due to processing dense frames individually. In this work, we propose a framework called Recurrent Residual Module (RRM) to accelerate the CNN inference for video recognition tasks. This framework has a novel design of using the similarity of the intermediate feature maps of two consecutive frames, to largely reduce the redundant computation. One unique property of the proposed method compared to previous work is that feature maps of each frame are precisely computed. The experiments show that, while maintaining the similar recognition performance, our RRM yields averagely 2× acceleration on the commonly used CNNs such as AlexNet, ResNet, deep compression model (thus 8-12× faster than the original dense models using the efficient inference engine), and impressively 9× acceleration on some binary networks such as XNOR-Nets (thus 500× faster than the original model). We further verify the effectiveness of the RRM on speeding up CNNs for video pose estimation and video object detection.

Currently, fusing visible and thermal images for facial expression recognition requires two modalities during both training and testing. Visible cameras are commonly used in real-life applications, and thermal cameras are typically only available in lab situations due to their high price. Thermal imaging for facial expression recognition is not frequently used in real-world situations. To address this, we propose a novel thermally enhanced facial expression recognition method which uses thermal images as privileged information to construct better visible feature representation and improved classifiers by incorporating adversarial learning and similarity constraints during training. Specifically, we train two deep neural networks from visible images and thermal images. We impose adversarial loss to enforce statistical similarity between the learned representations of two modalities, and a similarity constraint to regulate the mapping functions from visible and thermal representation to expressions. Thus, thermal images are leveraged to simultaneously improve visible feature representation and classification during training. To mimic real-world scenarios, only visible images are available during testing. We further extend the proposed expression recognition method for partially unpaired data to explore thermal images' supplementary role in visible facial expression recognition when visible images and thermal images are not synchronously recorded. Experimental results on the MAHNOB Laughter database demonstrate that our proposed method can effectively regularize visible representation and expression classifiers with the help of thermal images, achieving state-of-the-art recognition performance.

Visible facial images provide geometric and appearance patterns of facial expressions and are sensitive to illumination changes. Thermal facial images record facial temperature distribution and are robust to light conditions. Therefore, expression recognition is enhanced by visible and thermal image fusion. In most cases, only visible images are available due to the widespread popularity of visible cameras and the high cost of thermal cameras. Thus, we propose a novel visible expression recognition method by using thermal infrared (IR) data as privileged information, which is only available during training. Specifically, we first learn a deep model for visible images and thermal images. Then we use the learned feature representations to train support vector machine (SVM) classifiers for expression classification. We jointly refine the deep models as well as the SVM classifiers for both thermal images and visible images by imposing the constraint that the outputs of the SVM classifiers from two views are similar. Thermal IR images during training are then exploited to construct better facial representations and expression classifiers from visible images. We extend the proposed thermal augmented expression recognition method for partially unpaired data, acknowledging that visible images and thermal images maybe not be recorded synchronously. Experimental resulton the MAHNOB laughter database demonstrate that the proposed thermal augmented expression recognition method can effectively exploit thermal IR images' supplementary role for visible facial expression recognition during training to obtain better facial representations and a better visible expression classifier. The proposed thermal augmented expression recognition method achieves state-of-the-art expression recognition performance for both paired and unpaired facial images.

Current work on facial action unit (AU) recognition requires AU-labeled facial images. Although large amounts of facial images are readily available, AU annotation is expensive and time consuming. To address this, we propose a deep facial action unit recognition approach learning from partially AU-labeled data. The proposed approach makes full use of both partly available ground-truth AU labels and the readily available large scale facial images without annotation. Specifically, we propose to learn label distribution from the ground-truth AU labels, and then train the AU classifiers from the large-scale facial images by maximizing the log likelihood of the mapping functions of AUs with regard to the learnt label distribution for all training data and minimizing the error between predicted AUs and ground-truth AUs for labeled data simultaneously. A restricted Boltzmann machine is adopted to model AU label distribution, a deep neural network is used to learn facial representation from facial images, and the support vector machine is employed as the classifier. Experiments on two benchmark databases demonstrate the effectiveness of the proposed approach.

This paper proposes a novel deep two-view approach to learn features from both visible and thermal images and leverage the commonality among visible and thermal images for facial expression recognition from visible images. The thermal images are used as privileged information, which is required only during training to help visible images learn better features and classifier. Specifically, we first learn a deep model for visible images and thermal images respectively, and use the learned feature representations to train SVM classifiers for expression classification. We then jointly refine the deep models as well as the SVM classifiers for both thermal images and visible images by imposing the constraint that the outputs of the SVM classifiers from two views are similar. Therefore, the resulting representations and classifiers capture the inherent connections among visible facial image, infrared facial image and target expression labels, and hence improve the recognition performance for facial expression recognition from visible images during testing. Experimental results on the benchmark expression database demonstrate the effectiveness of our proposed method.