Search

Scene coordinate regression (SCR) methods have emerged as a promising area of research due to their potential for accurate visual localization. However, many existing SCR approaches train on samples from all image regions, including dynamic objects and texture-less areas. Utilizing these areas for optimization during training can potentially hamper the overall performance and efficiency of the model. In this study, we first perform an in-depth analysis to validate the adverse impacts of these areas. Drawing inspiration from our analysis, we then introduce an error-guided feature selection (EGFS) mechanism, in tandem with the use of the Segment Anything Model (SAM). This mechanism seeds low reprojection areas as prompts and expands them into error-guided masks, and then utilizes these masks to sample points and filter out problematic areas in an iterative manner. The experiments demonstrate that our method outperforms existing SCR approaches that do not rely on 3D information on the Cambridge Landmarks and Indoor6 datasets., Comment: ECCV2024

In this work, we report a pronounced light upconversion in few-layer transition metal dichalcogenides. Our joint theory-experiment study attributes the upconversion photoluminescence to a resonant exciton-exciton annihilation involving a pair of dark excitons with opposite momenta, followed by the spontaneous emission of upconverted bright excitons, which can have a high upconversion efficiency. Additionally, the upconversion photoluminescence is generic in MoS2, MoSe2, WS2, and WSe2, showing a high tuneability from green to ultraviolet light.

Kerr-Newman black holes in a de Sitter space have the limit of rotating Nariai black holes with the near-horizon geometry of a warped ${\rm dS}_3 \times {\rm S}^1/Z_2$ when the black hole horizon and the cosmological horizon coincide or approach close to each other. We study the effect of rotation on the emission of charges in the near-extremal rotating charged Nariai black hole and compare it to those from the near-extremal Nariai black hole and near-extremal Kerr-Newman black hole in de Sitter space. The emission has an exponential amplification for charges with high energy and becomes catastrophic when the two horizons are very close to each together. The angular momentum of black holes decreases the mean number of charges by a factor not by an order. We observe a catastrophic emission of boson condensation for charges with the effective energy equal to the chemical potential in the spacelike outer region of the cosmological horizon. Further, the rotating charged Nariai black holes can evolve into singular spacetimes with a naked singularity by the Schwinger pair production., Comment: 11 pages, 6 figures. arXiv admin note: text overlap with arXiv:2309.00218

Deep learning techniques have shown promising results in the automatic classification of respiratory sounds. However, accurately distinguishing these sounds in real-world noisy conditions poses challenges for clinical deployment. Additionally, predicting signals with only background noise could undermine user trust in the system. In this study, we propose an audio enhancement (AE) pipeline as a pre-processing step before respiratory sound classification, aiming to improve performance in noisy environments. Multiple experiments were conducted using different audio enhancement model structures, demonstrating improved classification performance compared to the baseline method of noise injection data augmentation. Specifically, the integration of the AE pipeline resulted in a 2.59% increase in the ICBHI classification score on the ICBHI respiratory sound dataset and a 2.51% improvement on our recently collected Formosa Archive of Breath Sounds (FABS) in multi-class noisy scenarios. Furthermore, a physician validation study assessed the clinical utility of our system. Quantitative analysis revealed enhancements in efficiency, diagnostic confidence, and trust during model-assisted diagnosis with our system compared to raw noisy recordings. Workflows integrating enhanced audio led to an 11.61% increase in diagnostic sensitivity and facilitated high-confidence diagnoses. Our findings demonstrate that incorporating an audio enhancement algorithm significantly enhances robustness and clinical utility., Comment: The following article has been submitted to The Journal of the Acoustical Society of America (JASA). After it is published, it will be found at https://pubs.aip.org/asa/jasa

As a fundamental structural feature, the symmetry of materials determines the exotic quantum properties in transition metal dichalcogenides (TMDs) with charge density wave (CDW). Breaking the inversion symmetry, the Janus structure, an artificially constructed lattice, provides an opportunity to tune the CDW states and the related properties. However, limited by the difficulties in atomic-level fabrication and material stability, the experimental visualization of the CDW states in 2D TMDs with Janus structure is still rare. Here, using surface selenization of VTe2, we fabricated monolayer Janus VTeSe. With scanning tunneling microscopy, an unusual root13-root13 CDW state with threefold rotational symmetry breaking was observed and characterized. Combined with theoretical calculations, we find this CDW state can be attributed to the charge modulation in the Janus VTeSe, beyond the conventional electron-phonon coupling. Our findings provide a promising platform for studying the CDW states and artificially tuning the electronic properties toward the applications.

While large language models (LLMs) such as Llama-2 or GPT-4 have shown impressive zero-shot performance, fine-tuning is still necessary to enhance their performance for customized datasets, domain-specific tasks, or other private needs. However, fine-tuning all parameters of LLMs requires significant hardware resources, which can be impractical for typical users. Therefore, parameter-efficient fine-tuning such as LoRA have emerged, allowing users to fine-tune LLMs without the need for considerable computing resources, with little performance degradation compared to fine-tuning all parameters. Unfortunately, recent studies indicate that fine-tuning can increase the risk to the safety of LLMs, even when data does not contain malicious content. To address this challenge, we propose Safe LoRA, a simple one-liner patch to the original LoRA implementation by introducing the projection of LoRA weights from selected layers to the safety-aligned subspace, effectively reducing the safety risks in LLM fine-tuning while maintaining utility. It is worth noting that Safe LoRA is a training-free and data-free approach, as it only requires the knowledge of the weights from the base and aligned LLMs. Our extensive experiments demonstrate that when fine-tuning on purely malicious data, Safe LoRA retains similar safety performance as the original aligned model. Moreover, when the fine-tuning dataset contains a mixture of both benign and malicious data, Safe LoRA mitigates the negative effect made by malicious data while preserving performance on downstream tasks.

Multivariate time series (MTS) data, when sampled irregularly and asynchronously, often present extensive missing values. Conventional methodologies for MTS analysis tend to rely on temporal embeddings based on timestamps that necessitate subsequent imputations, yet these imputed values frequently deviate substantially from their actual counterparts, thereby compromising prediction accuracy. Furthermore, these methods typically fail to provide robust initial embeddings for values infrequently observed or even absent within the training set, posing significant challenges to model generalizability. In response to these challenges, we propose SCAlable Numerical Embedding (SCANE), a novel framework that treats each feature value as an independent token, effectively bypassing the need for imputation. SCANE regularizes the traits of distinct feature embeddings and enhances representational learning through a scalable embedding mechanism. Coupling SCANE with the Transformer Encoder architecture, we develop the Scalable nUMerical eMbeddIng Transformer (SUMMIT), which is engineered to deliver precise predictive outputs for MTS characterized by prevalent missing entries. Our experimental validation, conducted across three disparate electronic health record (EHR) datasets marked by elevated missing value frequencies, confirms the superior performance of SUMMIT over contemporary state-of-the-art approaches addressing similar challenges. These results substantiate the efficacy of SCANE and SUMMIT, underscoring their potential applicability across a broad spectrum of MTS data analytical tasks.

Remarkable strides have been made in reconstructing static scenes or human bodies from monocular videos. Yet, the two problems have largely been approached independently, without much synergy. Most visual SLAM methods can only reconstruct camera trajectories and scene structures up to scale, while most HMR methods reconstruct human meshes in metric scale but fall short in reasoning with cameras and scenes. This work introduces Synergistic Camera and Human Reconstruction (SynCHMR) to marry the best of both worlds. Specifically, we design Human-aware Metric SLAM to reconstruct metric-scale camera poses and scene point clouds using camera-frame HMR as a strong prior, addressing depth, scale, and dynamic ambiguities. Conditioning on the dense scene recovered, we further learn a Scene-aware SMPL Denoiser to enhance world-frame HMR by incorporating spatio-temporal coherency and dynamic scene constraints. Together, they lead to consistent reconstructions of camera trajectories, human meshes, and dense scene point clouds in a common world frame. Project page: https://paulchhuang.github.io/synchmr, Comment: CVPR 2024

Conventional Federated Learning (FL) involves collaborative training of a global model while maintaining user data privacy. One of its branches, decentralized FL, is a serverless network that allows clients to own and optimize different local models separately, which results in saving management and communication resources. Despite the promising advancements in decentralized FL, it may reduce model generalizability due to lacking a global model. In this scenario, managing data and model heterogeneity among clients becomes a crucial problem, which poses a unique challenge that must be overcome: How can every client's local model learn generalizable representation in a decentralized manner? To address this challenge, we propose a novel Decentralized FL technique by introducing Synthetic Anchors, dubbed as DeSA. Based on the theory of domain adaptation and Knowledge Distillation (KD), we theoretically and empirically show that synthesizing global anchors based on raw data distribution facilitates mutual knowledge transfer. We further design two effective regularization terms for local training: 1) REG loss that regularizes the distribution of the client's latent embedding with the anchors and 2) KD loss that enables clients to learn from others. Through extensive experiments on diverse client data distributions, we showcase the effectiveness of DeSA in enhancing both inter- and intra-domain accuracy of each client., Comment: Paper Accepted at ICML 2024, 23 pages

We present MatAtlas, a method for consistent text-guided 3D model texturing. Following recent progress we leverage a large scale text-to-image generation model (e.g., Stable Diffusion) as a prior to texture a 3D model. We carefully design an RGB texturing pipeline that leverages a grid pattern diffusion, driven by depth and edges. By proposing a multi-step texture refinement process, we significantly improve the quality and 3D consistency of the texturing output. To further address the problem of baked-in lighting, we move beyond RGB colors and pursue assigning parametric materials to the assets. Given the high-quality initial RGB texture, we propose a novel material retrieval method capitalized on Large Language Models (LLM), enabling editabiliy and relightability. We evaluate our method on a wide variety of geometries and show that our method significantly outperform prior arts. We also analyze the role of each component through a detailed ablation study.

Generating video background that tailors to foreground subject motion is an important problem for the movie industry and visual effects community. This task involves synthesizing background that aligns with the motion and appearance of the foreground subject, while also complies with the artist's creative intention. We introduce ActAnywhere, a generative model that automates this process which traditionally requires tedious manual efforts. Our model leverages the power of large-scale video diffusion models, and is specifically tailored for this task. ActAnywhere takes a sequence of foreground subject segmentation as input and an image that describes the desired scene as condition, to produce a coherent video with realistic foreground-background interactions while adhering to the condition frame. We train our model on a large-scale dataset of human-scene interaction videos. Extensive evaluations demonstrate the superior performance of our model, significantly outperforming baselines. Moreover, we show that ActAnywhere generalizes to diverse out-of-distribution samples, including non-human subjects. Please visit our project webpage at https://actanywhere.github.io.

Modeling the interaction between humans and objects has been an emerging research direction in recent years. Capturing human-object interaction is however a very challenging task due to heavy occlusion and complex dynamics, which requires understanding not only 3D human pose, and object pose but also the interaction between them. Reconstruction of 3D humans and objects has been two separate research fields in computer vision for a long time. We hence proposed the first RHOBIN challenge: reconstruction of human-object interactions in conjunction with the RHOBIN workshop. It was aimed at bringing the research communities of human and object reconstruction as well as interaction modeling together to discuss techniques and exchange ideas. Our challenge consists of three tracks of 3D reconstruction from monocular RGB images with a focus on dealing with challenging interaction scenarios. Our challenge attracted more than 100 participants with more than 300 submissions, indicating the broad interest in the research communities. This paper describes the settings of our challenge and discusses the winning methods of each track in more detail. We observe that the human reconstruction task is becoming mature even under heavy occlusion settings while object pose estimation and joint reconstruction remain challenging tasks. With the growing interest in interaction modeling, we hope this report can provide useful insights and foster future research in this direction. Our workshop website can be found at \href{https://rhobin-challenge.github.io/}{https://rhobin-challenge.github.io/}., Comment: 14 pages, 5 tables, 7 figure. Technical report of the CVPR'23 workshop: RHOBIN challenge (https://rhobin-challenge.github.io/)

In recent years, foundation models (FMs) have solidified their role as cornerstone advancements in the deep learning domain. By extracting intricate patterns from vast datasets, these models consistently achieve state-of-the-art results across a spectrum of downstream tasks, all without necessitating extensive computational resources. Notably, MedCLIP, a vision-language contrastive learning-based medical FM, has been designed using unpaired image-text training. While the medical domain has often adopted unpaired training to amplify data, the exploration of potential security concerns linked to this approach hasn't kept pace with its practical usage. Notably, the augmentation capabilities inherent in unpaired training also indicate that minor label discrepancies can result in significant model deviations. In this study, we frame this label discrepancy as a backdoor attack problem. We further analyze its impact on medical FMs throughout the FM supply chain. Our evaluation primarily revolves around MedCLIP, emblematic of medical FM employing the unpaired strategy. We begin with an exploration of vulnerabilities in MedCLIP stemming from unpaired image-text matching, termed BadMatch. BadMatch is achieved using a modest set of wrongly labeled data. Subsequently, we disrupt MedCLIP's contrastive learning through BadDist-assisted BadMatch by introducing a Bad-Distance between the embeddings of clean and poisoned data. Additionally, combined with BadMatch and BadDist, the attacking pipeline consistently fends off backdoor assaults across diverse model designs, datasets, and triggers. Also, our findings reveal that current defense strategies are insufficient in detecting these latent threats in medical FMs' supply chains., Comment: Paper Accepted at the 2nd IEEE Conference on Secure and Trustworthy Machine Learning

Wind turbines are subjected to continuous rotational stresses and unusual external forces such as storms, lightning, strikes by flying objects, etc., which may cause defects in turbine blades. Hence, it requires a periodical inspection to ensure proper functionality and avoid catastrophic failure. The task of inspection is challenging due to the remote location and inconvenient reachability by human inspection. Researchers used images with cropped defects from the wind turbine in the literature. They neglected possible background biases, which may hinder real-time and autonomous defect detection using aerial vehicles such as drones or others. To overcome such challenges, in this paper, we experiment with defect detection accuracy by having the defects with the background using a two-step deep-learning methodology. In the first step, we develop virtual models of wind turbines to synthesize the near-reality images for four types of common defects - cracks, leading edge erosion, bending, and light striking damage. The Unity perception package is used to generate wind turbine blade defects images with variations in background, randomness, camera angle, and light effects. In the second step, a customized U-Net architecture is trained to classify and segment the defect in turbine blades. The outcomes of U-Net architecture have been thoroughly tested and compared with 5-fold validation datasets. The proposed methodology provides reasonable defect detection accuracy, making it suitable for autonomous and remote inspection through aerial vehicles.

The extremely large-scale massive multiple-input multiple-output (XL-MIMO) has the potential to achieve boosted spectral efficiency and refined spatial resolution for future wireless networks. However, channel estimation for XL-MIMO is challenging since the large number of antennas results in high computational complexity with the near-field effect. In this letter, we propose a low-complexity sequential angle-distance channel estimation (SADCE) method for near-field XL-MIMO systems equipped with uniformly planar arrays (UPA). Specifically, we first successfully decouple the angle and distance parameters, which allows us to devise a two-dimensional discrete Fourier transform (2D-DFT) method for angle parameters estimation. Then, a low-complexity distance estimation method is proposed with a closed-form solution. Compared with existing methods, the proposed method achieves significant performance gain with noticeably reduced computational complexity.Numerical results verify the superiority of the proposed near-field channel estimation algorithm.

Traditional 3D content creation tools empower users to bring their imagination to life by giving them direct control over a scene's geometry, appearance, motion, and camera path. Creating computer-generated videos, however, is a tedious manual process, which can be automated by emerging text-to-video diffusion models. Despite great promise, video diffusion models are difficult to control, hindering a user to apply their own creativity rather than amplifying it. To address this challenge, we present a novel approach that combines the controllability of dynamic 3D meshes with the expressivity and editability of emerging diffusion models. For this purpose, our approach takes an animated, low-fidelity rendered mesh as input and injects the ground truth correspondence information obtained from the dynamic mesh into various stages of a pre-trained text-to-image generation model to output high-quality and temporally consistent frames. We demonstrate our approach on various examples where motion can be obtained by animating rigged assets or changing the camera path., Comment: Project page: https://primecai.github.io/generative_rendering/

Diffusion models for text-to-image (T2I) synthesis, such as Stable Diffusion (SD), have recently demonstrated exceptional capabilities for generating high-quality content. However, this progress has raised several concerns of potential misuse, particularly in creating copyrighted, prohibited, and restricted content, or NSFW (not safe for work) images. While efforts have been made to mitigate such problems, either by implementing a safety filter at the evaluation stage or by fine-tuning models to eliminate undesirable concepts or styles, the effectiveness of these safety measures in dealing with a wide range of prompts remains largely unexplored. In this work, we aim to investigate these safety mechanisms by proposing one novel concept retrieval algorithm for evaluation. We introduce Ring-A-Bell, a model-agnostic red-teaming tool for T2I diffusion models, where the whole evaluation can be prepared in advance without prior knowledge of the target model. Specifically, Ring-A-Bell first performs concept extraction to obtain holistic representations for sensitive and inappropriate concepts. Subsequently, by leveraging the extracted concept, Ring-A-Bell automatically identifies problematic prompts for diffusion models with the corresponding generation of inappropriate content, allowing the user to assess the reliability of deployed safety mechanisms. Finally, we empirically validate our method by testing online services such as Midjourney and various methods of concept removal. Our results show that Ring-A-Bell, by manipulating safe prompting benchmarks, can transform prompts that were originally regarded as safe to evade existing safety mechanisms, thus revealing the defects of the so-called safety mechanisms which could practically lead to the generation of harmful contents. Our codes are available at https://github.com/chiayi-hsu/Ring-A-Bell., Comment: This paper has already been accepted by ICLR 2024. This version is the camera-ready version

The challenge of navigation in environments with dynamic objects continues to be a central issue in the study of autonomous agents. While predictive methods hold promise, their reliance on precise state information makes them less practical for real-world implementation. This study presents visual forecasting as an innovative alternative. By introducing intuitive visual cues, this approach projects the future trajectories of dynamic objects to improve agent perception and enable anticipatory actions. Our research explores two distinct strategies for conveying predictive information through visual forecasting: (1) sequences of bounding boxes, and (2) augmented paths. To validate the proposed visual forecasting strategies, we initiate evaluations in simulated environments using the Unity engine and then extend these evaluations to real-world scenarios to assess both practicality and effectiveness. The results confirm the viability of visual forecasting as a promising solution for navigation and obstacle avoidance in dynamic environments., Comment: Tsung-Chih Chiang, Ting-Ru Liu, Chun-Wei Huang, and Jou-Min Liu contributed equally to this work; This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

We propose an extended compass model that hosts subsystem symmetries and has potential experimental relevance with 3d transition metal compounds. The subsystem symmetries strongly constrain the mobility of spin excitations and lead to profound consequences. At the quantum critical point we find the presence of "critical Bose surface" along the entire $k_x$ and $k_y$ axis. Across which we find a nodal-line spin liquid that undergoes nematic instability at low temperatures. In the ferro-quadrupole phase, we find that one excitation is immobile individually analogous to "fractons"., Comment: 14 pages, 7 figures

Ferroelectricity, a spontaneous and reversible electric polarization, is found in certain classes of van der Waals (vdW) material heterostructures. The discovery of ferroelectricity in twisted vdW layers provides new opportunities to engineer spatially dependent electric and optical properties associated with the configuration of moir\'e superlattice domains and the network of domain walls. Here, we employ near-field infrared nano-imaging and nano-photocurrent measurements to study ferroelectricity in minimally twisted WSe2. The ferroelectric domains are visualized through the imaging of the plasmonic response in a graphene monolayer adjacent to the moir\'e WSe2 bilayers. Specifically, we find that the ferroelectric polarization in moir\'e domains is imprinted on the plasmonic response of the graphene. Complementary nano-photocurrent measurements demonstrate that the optoelectronic properties of graphene are also modulated by the proximal ferroelectric domains. Our approach represents an alternative strategy for studying moir\'e ferroelectricity at native length scales and opens promising prospects for (opto)electronic devices., Comment: 19 pages, 3 figures

Morphable models are fundamental to numerous human-centered processes as they offer a simple yet expressive shape space. Creating such morphable models, however, is both tedious and expensive. The main challenge is establishing dense correspondences across raw scans that capture sufficient shape variation. This is often addressed using a mix of significant manual intervention and non-rigid registration. We observe that creating a shape space and solving for dense correspondence are tightly coupled -- while dense correspondence is needed to build shape spaces, an expressive shape space provides a reduced dimensional space to regularize the search. We introduce BLiSS, a method to solve both progressively. Starting from a small set of manually registered scans to bootstrap the process, we enrich the shape space and then use that to get new unregistered scans into correspondence automatically. The critical component of BLiSS is a non-linear deformation model that captures details missed by the low-dimensional shape space, thus allowing progressive enrichment of the space., Comment: 12 pages, 10 figures

Using both the in-out formalism and the monodromy method, we study the emission of charges from near-extremal charged Nariai black holes with the black hole and cosmological horizons close to each other, whose near-horizon geometry is $\mathrm{dS}_2 \times \mathrm{S}^2$. The emission becomes catastrophic for a charge with energy greater than its chemical potential, whose leading exponential factor increases inversely proportional to the separation of two horizons. This implies that near-extremal Nariai black holes with large charges quickly evaporate dominantly through the charge emission and evolve to black holes with a naked singularity, in contrast to near-extremal RN-dS black holes that have the Breitenlohner-Friedman bound below which they become stable against Hawking radiation and Schwinger effect of charge emission. The near-extremal Nariai black holes with small charges, which are close to near-extremal Schwarzschild-dS black holes, emit dominantly charge-neutral particles and evolve to black holes with increasing charge to mass ratio. We illuminate the origin of the catastrophic emission in the phase-integral formulation and monodromy method by comparing near-extremal charged Nariai black holes with near-extremal RN-dS black holes., Comment: revised version including corrections of mistakes and more physical discussions

Given a gene-tree labeled topology $G$ and a species tree $S$, the "ancestral configurations" at an internal node $k$ of $S$ represent the combinatorially different sets of gene lineages that can be present at $k$ when all possible realizations of $G$ in $S$ are considered. Ancestral configurations have been introduced as a data structure for evaluating the conditional probability of a gene-tree labeled topology given a species tree, and their enumeration assists in describing the complexity of this computation. In the case that the gene-tree labeled topology $G=t$ matches that of the species tree $S$, by techniques of analytic combinatorics, we study distributional properties of the "total" number of ancestral configurations measured across the different nodes of a random labeled topology $t$ selected under the uniform and the Yule probability models. Under both of these probabilistic scenarios, we show that the total number $T_n$ of ancestral configurations of a random labeled topology of $n$ taxa asymptotically follows a lognormal distribution. Over uniformly distributed labeled topologies, the asymptotic growth of the mean and the variance of $T_n$ are found to satisfy $\mathbb{E}_{\rm U}[T_n] \sim 2.449 \cdot 1.333^n$ and $\mathbb{V}_{\rm U}[T_n] \sim 5.050 \cdot 1.822^n$, respectively. Under the Yule model, which assigns higher probabilities to more balanced labeled topologies, we obtain the mean $\mathbb{E}_{\rm Y}[T_n] \sim 1.425^n$ and the variance $\mathbb{V}_{\rm Y}[T_n] \sim 2.045^n$.

Existing research on data valuation in federated and swarm learning focuses on valuing client contributions and works best when data across clients is independent and identically distributed (IID). In practice, data is rarely distributed IID. We develop an approach called DDVal for decentralized data valuation, capable of valuing individual data points in federated and swarm learning. DDVal is based on sharing deep features and approximating Shapley values through a k-nearest neighbor approximation method. This allows for novel applications, for example, to simultaneously reward institutions and individuals for providing data to a decentralized machine learning task. The valuation of data points through DDVal allows to also draw hierarchical conclusions on the contribution of institutions, and we empirically show that the accuracy of DDVal in estimating institutional contributions is higher than existing Shapley value approximation methods for federated learning. Specifically, it reaches a cosine similarity in approximating Shapley values of 99.969 % in both, IID and non-IID data distributions across institutions, compared with 99.301 % and 97.250 % for the best state of the art methods. DDVal scales with the number of data points instead of the number of clients, and has a loglinear complexity. This scales more favorably than existing approaches with an exponential complexity. We show that DDVal is especially efficient in data distribution scenarios with many clients that have few data points - for example, more than 16 clients with 8,000 data points each. By integrating DDVal into a decentralized system, we show that it is not only suitable for centralized federated learning, but also decentralized swarm learning, which aligns well with the research on emerging internet technologies such as web3 to reward users for providing data to algorithms.

Differentially private synthetic data is a promising alternative for sensitive data release. Many differentially private generative models have been proposed in the literature. Unfortunately, they all suffer from the low utility of the synthetic data, particularly for images of high resolutions. Here, we propose DPAF, an effective differentially private generative model for high-dimensional image synthesis. Different from the prior private stochastic gradient descent-based methods that add Gaussian noises in the backward phase during the model training, DPAF adds a differentially private feature aggregation in the forward phase, bringing advantages, including the reduction of information loss in gradient clipping and low sensitivity for the aggregation. Moreover, as an improper batch size has an adverse impact on the utility of synthetic data, DPAF also tackles the problem of setting a proper batch size by proposing a novel training strategy that asymmetrically trains different parts of the discriminator. We extensively evaluate different methods on multiple image datasets (up to images of 128x128 resolution) to demonstrate the performance of DPAF.

Sign language (SL) is the primary method of communication for the 70 million Deaf people around the world. Video dictionaries of isolated signs are a core SL learning tool. Replacing these with 3D avatars can aid learning and enable AR/VR applications, improving access to technology and online media. However, little work has attempted to estimate expressive 3D avatars from SL video; occlusion, noise, and motion blur make this task difficult. We address this by introducing novel linguistic priors that are universally applicable to SL and provide constraints on 3D hand pose that help resolve ambiguities within isolated signs. Our method, SGNify, captures fine-grained hand pose, facial expression, and body movement fully automatically from in-the-wild monocular SL videos. We evaluate SGNify quantitatively by using a commercial motion-capture system to compute 3D avatars synchronized with monocular video. SGNify outperforms state-of-the-art 3D body-pose- and shape-estimation methods on SL videos. A perceptual study shows that SGNify's 3D reconstructions are significantly more comprehensible and natural than those of previous methods and are on par with the source videos. Code and data are available at $\href{http://sgnify.is.tue.mpg.de}{\text{sgnify.is.tue.mpg.de}}$.