Search

Recent image-to-3D reconstruction models have greatly advanced geometry generation, but they still struggle to faithfully generate realistic appearance. To address this, we introduce ARM, a novel method that reconstructs high-quality 3D meshes and realistic appearance from sparse-view images. The core of ARM lies in decoupling geometry from appearance, processing appearance within the UV texture space. Unlike previous methods, ARM improves texture quality by explicitly back-projecting measurements onto the texture map and processing them in a UV space module with a global receptive field. To resolve ambiguities between material and illumination in input images, ARM introduces a material prior that encodes semantic appearance information, enhancing the robustness of appearance decomposition. Trained on just 8 H100 GPUs, ARM outperforms existing methods both quantitatively and qualitatively.

Large language models (LLMs), with advanced linguistic capabilities, have been employed in reranking tasks through a sequence-to-sequence approach. In this paradigm, multiple passages are reranked in a listwise manner and a textual reranked permutation is generated. However, due to the limited context window of LLMs, this reranking paradigm requires a sliding window strategy to iteratively handle larger candidate sets. This not only increases computational costs but also restricts the LLM from fully capturing all the comparison information for all candidates. To address these challenges, we propose a novel self-calibrated listwise reranking method, which aims to leverage LLMs to produce global relevance scores for ranking. To achieve it, we first propose the relevance-aware listwise reranking framework, which incorporates explicit list-view relevance scores to improve reranking efficiency and enable global comparison across the entire candidate set. Second, to ensure the comparability of the computed scores, we propose self-calibrated training that uses point-view relevance assessments generated internally by the LLM itself to calibrate the list-view relevance assessments. Extensive experiments and comprehensive analysis on the BEIR benchmark and TREC Deep Learning Tracks demonstrate the effectiveness and efficiency of our proposed method.

Video Multimodal Large Language Models (MLLMs) have shown remarkable capability of understanding the video semantics on various downstream tasks. Despite the advancements, there is still a lack of systematic research on visual context representation, which refers to the scheme to select frames from a video and further select the tokens from a frame. In this paper, we explore the design space for visual context representation, and aim to improve the performance of video MLLMs by finding more effective representation schemes. Firstly, we formulate the task of visual context representation as a constrained optimization problem, and model the language modeling loss as a function of the number of frames and the number of embeddings (or tokens) per frame, given the maximum visual context window size. Then, we explore the scaling effects in frame selection and token selection respectively, and fit the corresponding function curve by conducting extensive empirical experiments. We examine the effectiveness of typical selection strategies and present empirical findings to determine the two factors. Furthermore, we study the joint effect of frame selection and token selection, and derive the optimal formula for determining the two factors. We demonstrate that the derived optimal settings show alignment with the best-performed results of empirical experiments. Our code and model are available at: https://github.com/RUCAIBox/Opt-Visor., Comment: Long Video MLLM; work in progress

We present a spatial and angular Gaussian based representation and a triple splatting process, for real-time, high-quality novel lighting-and-view synthesis from multi-view point-lit input images. To describe complex appearance, we employ a Lambertian plus a mixture of angular Gaussians as an effective reflectance function for each spatial Gaussian. To generate self-shadow, we splat all spatial Gaussians towards the light source to obtain shadow values, which are further refined by a small multi-layer perceptron. To compensate for other effects like global illumination, another network is trained to compute and add a per-spatial-Gaussian RGB tuple. The effectiveness of our representation is demonstrated on 30 samples with a wide variation in geometry (from solid to fluffy) and appearance (from translucent to anisotropic), as well as using different forms of input data, including rendered images of synthetic/reconstructed objects, photographs captured with a handheld camera and a flash, or from a professional lightstage. We achieve a training time of 40-70 minutes and a rendering speed of 90 fps on a single commodity GPU. Our results compare favorably with state-of-the-art techniques in terms of quality/performance. Our code and data are publicly available at https://GSrelight.github.io/., Comment: Accepted to SIGGRAPH Asia 2024. Project page: https://gsrelight.github.io/

Multi-lingual ability transfer has become increasingly important for the broad application of large language models (LLMs). Existing work highly relies on training with the multi-lingual ability-related data, which may be not available for low-resource languages. To solve it, we propose a Multi-lingual Ability Extraction and Transfer approach, named as MAET. Our key idea is to decompose and extract language-agnostic ability-related weights from LLMs, and transfer them across different languages by simple addition and subtraction operations without training. Specially, our MAET consists of the extraction and transfer stages. In the extraction stage, we firstly locate key neurons that are highly related to specific abilities, and then employ them to extract the transferable ability-specific weights. In the transfer stage, we further select the ability-related parameter tensors, and design the merging strategy based on the linguistic and ability specific weights, to build the multi-lingual ability-enhanced LLM. To demonstrate the effectiveness of our proposed approach, we conduct extensive experiments on mathematical and scientific tasks in both high-resource lingual and low-resource lingual scenarios. Experiment results have shown that MAET can effectively and efficiently extract and transfer the advanced abilities, and outperform training-based baseline methods. Our code and data are available at \url{https://github.com/RUCAIBox/MAET}., Comment: 18 Pages. Working in progress

Current co-speech motion generation approaches usually focus on upper body gestures following speech contents only, while lacking supporting the elaborate control of synergistic full-body motion based on text prompts, such as talking while walking. The major challenges lie in 1) the existing speech-to-motion datasets only involve highly limited full-body motions, making a wide range of common human activities out of training distribution; 2) these datasets also lack annotated user prompts. To address these challenges, we propose SynTalker, which utilizes the off-the-shelf text-to-motion dataset as an auxiliary for supplementing the missing full-body motion and prompts. The core technical contributions are two-fold. One is the multi-stage training process which obtains an aligned embedding space of motion, speech, and prompts despite the significant distributional mismatch in motion between speech-to-motion and text-to-motion datasets. Another is the diffusion-based conditional inference process, which utilizes the separate-then-combine strategy to realize fine-grained control of local body parts. Extensive experiments are conducted to verify that our approach supports precise and flexible control of synergistic full-body motion generation based on both speeches and user prompts, which is beyond the ability of existing approaches., Comment: Project Page: https://robinwitch.github.io/SynTalker-Page

Large-scale text-guided image diffusion models have shown astonishing results in text-to-image (T2I) generation. However, applying these models to synthesize textures for 3D geometries remains challenging due to the domain gap between 2D images and textures on a 3D surface. Early works that used a projecting-and-inpainting approach managed to preserve generation diversity but often resulted in noticeable artifacts and style inconsistencies. While recent methods have attempted to address these inconsistencies, they often introduce other issues, such as blurring, over-saturation, or over-smoothing. To overcome these challenges, we propose a novel text-to-texture synthesis framework that leverages pretrained diffusion models. We first introduce a local attention reweighing mechanism in the self-attention layers to guide the model in concentrating on spatial-correlated patches across different views, thereby enhancing local details while preserving cross-view consistency. Additionally, we propose a novel latent space merge pipeline, which further ensures consistency across different viewpoints without sacrificing too much diversity. Our method significantly outperforms existing state-of-the-art techniques regarding texture consistency and visual quality, while delivering results much faster than distillation-based methods. Importantly, our framework does not require additional training or fine-tuning, making it highly adaptable to a wide range of models available on public platforms.

Current emotional text-to-speech (TTS) systems face challenges in mimicking a broad spectrum of human emotions due to the inherent complexity of emotions and limitations in emotional speech datasets and models. This paper proposes a TTS framework that facilitates control over pleasure, arousal, and dominance, and can synthesize a diversity of emotional styles without requiring any emotional speech data during TTS training. We train an emotional attribute predictor using only categorical labels from speech data, aligning with psychological research and incorporating anchored dimensionality reduction on self-supervised learning (SSL) features. The TTS framework converts text inputs into phonetic tokens via an autoregressive language model and uses pseudo-emotional dimensions to guide the parallel prediction of fine-grained acoustic details. Experiments conducted on the LibriTTS dataset demonstrate that our framework can synthesize speech with enhanced naturalness and a variety of emotional styles by effectively controlling emotional dimensions, even without the inclusion of any emotional speech during TTS training., Comment: submitted to ICASSP 2025

While current emotional text-to-speech (TTS) systems can generate highly intelligible emotional speech, achieving fine control over emotion rendering of the output speech still remains a significant challenge. In this paper, we introduce ParaEVITS, a novel emotional TTS framework that leverages the compositionality of natural language to enhance control over emotional rendering. By incorporating a text-audio encoder inspired by ParaCLAP, a contrastive language-audio pretraining (CLAP) model for computational paralinguistics, the diffusion model is trained to generate emotional embeddings based on textual emotional style descriptions. Our framework first trains on reference audio using the audio encoder, then fine-tunes a diffusion model to process textual inputs from ParaCLAP's text encoder. During inference, speech attributes such as pitch, jitter, and loudness are manipulated using only textual conditioning. Our experiments demonstrate that ParaEVITS effectively control emotion rendering without compromising speech quality. Speech demos are publicly available.

Ferroelasticity describes a phenomenon in which a material exhibits two or more equally stable orientation variants and can be switched from one form to another under an applied stress. Recent works have demonstrated that two-dimensional layered organic$-$inorganic hybrid Ruddlesden$-$Popper perovskites can serve as ideal platforms for realizing ferroelasticity, however, the ferroelastic (FE) behavior of structures with a single octahedra layer such as (BA)$_2$PbI$_4$ (BA = CH$_3$(CH$_2$)$_3$NH$_3$$^+$) has remained elusive. Herein, by using a combined first-principles and metadynamics approach, the FE behavior of (BA)$_2$PbI$_4$ under mechanical and thermal stresses is uncovered. FE switching is mediated by cross-plane intermolecular coupling, which could occur through multiple rotational modes, rendering the formation of FE domains and several metastable paraelastic (PE) phases. Such metastable phases are akin to wrinkled structures in other layered materials and can act as a "funnel" of hole carriers. Thermal excitation tends to flatten the kinetic barriers of the transition pathways between orientation variants, suggesting an enhanced concentration of metastable PE states at high temperatures, while halogen mixing with Br raises these barriers and conversely lowers the concentration of PE states. These findings reveal the rich structural diversity of (BA)$_2$PbI$_4$ domains, which can play a vital role in enhancing the optoelectronic properties of the perovskite and raise exciting prospects for mechanical switching, shape memory, and information processing.

Previous low-light image enhancement (LLIE) approaches, while employing frequency decomposition techniques to address the intertwined challenges of low frequency (e.g., illumination recovery) and high frequency (e.g., noise reduction), primarily focused on the development of dedicated and complex networks to achieve improved performance. In contrast, we reveal that an advanced disentanglement paradigm is sufficient to consistently enhance state-of-the-art methods with minimal computational overhead. Leveraging the image Laplace decomposition scheme, we propose a novel low-frequency consistency method, facilitating improved frequency disentanglement optimization. Our method, seamlessly integrating with various models such as CNNs, Transformers, and flow-based and diffusion models, demonstrates remarkable adaptability. Noteworthy improvements are showcased across five popular benchmarks, with up to 7.68dB gains on PSNR achieved for six state-of-the-art models. Impressively, our approach maintains efficiency with only 88K extra parameters, setting a new standard in the challenging realm of low-light image enhancement., Comment: Accepted to ECCV 2024, Github \url{https://github.com/redrock303/ADF-LLIE}

We propose progressive radiance distillation, an inverse rendering method that combines physically-based rendering with Gaussian-based radiance field rendering using a distillation progress map. Taking multi-view images as input, our method starts from a pre-trained radiance field guidance, and distills physically-based light and material parameters from the radiance field using an image-fitting process. The distillation progress map is initialized to a small value, which favors radiance field rendering. During early iterations when fitted light and material parameters are far from convergence, the radiance field fallback ensures the sanity of image loss gradients and avoids local minima that attracts under-fit states. As fitted parameters converge, the physical model gradually takes over and the distillation progress increases correspondingly. In presence of light paths unmodeled by the physical model, the distillation progress never finishes on affected pixels and the learned radiance field stays in the final rendering. With this designed tolerance for physical model limitations, we prevent unmodeled color components from leaking into light and material parameters, alleviating relighting artifacts. Meanwhile, the remaining radiance field compensates for the limitations of the physical model, guaranteeing high-quality novel views synthesis. Experimental results demonstrate that our method significantly outperforms state-of-the-art techniques quality-wise in both novel view synthesis and relighting. The idea of progressive radiance distillation is not limited to Gaussian splatting. We show that it also has positive effects for prominently specular scenes when adapted to a mesh-based inverse rendering method.

Continual pre-training (CPT) has been an important approach for adapting language models to specific domains or tasks. To make the CPT approach more traceable, this paper presents a technical report for continually pre-training Llama-3 (8B), which significantly enhances the Chinese language ability and scientific reasoning ability of the backbone model. To enhance the new abilities while retaining the original abilities, we design specific data mixture and curriculum strategies by utilizing existing datasets and synthesizing high-quality datasets. Specifically, we synthesize multidisciplinary scientific question and answer (QA) pairs based on related web pages, and subsequently incorporate these synthetic data to improve the scientific reasoning ability of Llama-3. We refer to the model after CPT as Llama-3-SynE (Synthetic data Enhanced Llama-3). We also present the tuning experiments with a relatively small model -- TinyLlama, and employ the derived findings to train the backbone model. Extensive experiments on a number of evaluation benchmarks show that our approach can largely improve the performance of the backbone models, including both the general abilities (+8.81 on C-Eval and +6.31 on CMMLU) and the scientific reasoning abilities (+12.00 on MATH and +4.13 on SciEval), without hurting the original capacities. Our model, data, and codes are available at https://github.com/RUC-GSAI/Llama-3-SynE., Comment: 16 pages, 10 figures, 16 tables

Creating relightable and animatable avatars from multi-view or monocular videos is a challenging task for digital human creation and virtual reality applications. Previous methods rely on neural radiance fields or ray tracing, resulting in slow training and rendering processes. By utilizing Gaussian Splatting, we propose a simple and efficient method to decouple body materials and lighting from sparse-view or monocular avatar videos, so that the avatar can be rendered simultaneously under novel viewpoints, poses, and lightings at interactive frame rates (6.9 fps). Specifically, we first obtain the canonical body mesh using a signed distance function and assign attributes to each mesh vertex. The Gaussians in the canonical space then interpolate from nearby body mesh vertices to obtain the attributes. We subsequently deform the Gaussians to the posed space using forward skinning, and combine the learnable environment light with the Gaussian attributes for shading computation. To achieve fast shadow modeling, we rasterize the posed body mesh from dense viewpoints to obtain the visibility. Our approach is not only simple but also fast enough to allow interactive rendering of avatar animation under environmental light changes. Experiments demonstrate that, compared to previous works, our method can render higher quality results at a faster speed on both synthetic and real datasets.

The fast growth of E-Commerce creates a global market worth USD 821 billion for online fashion shopping. What unique about fashion presentation is that, the same design can usually be offered with different cloths textures. However, only real video capturing or manual per-frame editing can be used for virtual showcase on the same design with different textures, both of which are heavily labor intensive. In this paper, we present a pattern-based approach for UV and shading recovery from a captured real video so that the garment's texture can be replaced automatically. The core of our approach is a per-pixel UV regression module via blended-weight multilayer perceptrons (MLPs) driven by the detected discrete correspondences from the cloth pattern. We propose a novel loss on the Jacobian of the UV mapping to create pleasant seams around the folding areas and the boundary of occluded regions while avoiding UV distortion. We also adopts the temporal constraint to ensure consistency and accuracy in UV prediction across adjacent frames. We show that our approach is robust to a variety type of clothes, in the wild illuminations and with challenging motions. We show plausible texture replacement results in our experiment, in which the folding and overlapping of the garment can be greatly preserved. We also show clear qualitative and quantitative improvement compared to the baselines as well. With the one-click setup, we look forward to our approach contributing to the growth of fashion E-commerce., Comment: Accepted to IEEE Transactions on Visualization and Computer Graphics

To facilitate the research on large language models (LLMs), this paper presents a comprehensive and unified library, LLMBox, to ease the development, use, and evaluation of LLMs. This library is featured with three main merits: (1) a unified data interface that supports the flexible implementation of various training strategies, (2) a comprehensive evaluation that covers extensive tasks, datasets, and models, and (3) more practical consideration, especially on user-friendliness and efficiency. With our library, users can easily reproduce existing methods, train new models, and conduct comprehensive performance comparisons. To rigorously test LLMBox, we conduct extensive experiments in a diverse coverage of evaluation settings, and experimental results demonstrate the effectiveness and efficiency of our library in supporting various implementations related to LLMs. The detailed introduction and usage guidance can be found at https://github.com/RUCAIBox/LLMBox., Comment: Accepted by ACL 2024 Demo

Large language models (LLMs) have become the foundation of many applications, leveraging their extensive capabilities in processing and understanding natural language. While many open-source LLMs have been released with technical reports, the lack of training details hinders further research and development. This paper presents the development of YuLan, a series of open-source LLMs with $12$ billion parameters. The base model of YuLan is pre-trained on approximately $1.7$T tokens derived from a diverse corpus, including massive English, Chinese, and multilingual texts. We design a three-stage pre-training method to enhance YuLan's overall capabilities. Subsequent phases of training incorporate instruction-tuning and human alignment, employing a substantial volume of high-quality synthesized data. To facilitate the learning of complex and long-tail knowledge, we devise a curriculum-learning framework throughout across these stages, which helps LLMs learn knowledge in an easy-to-hard manner. YuLan's training is finished on Jan, 2024 and has achieved performance on par with state-of-the-art LLMs across various English and Chinese benchmarks. This paper outlines a comprehensive technical roadmap for developing LLMs from scratch. Our model and codes are available at https://github.com/RUC-GSAI/YuLan-Chat.

With the rapid development of video Multimodal Large Language Models (MLLMs), numerous benchmarks have been proposed to assess their video understanding capability. However, due to the lack of rich events in the videos, these datasets may suffer from the short-cut bias that the answers can be deduced from a few frames, without the need to watch the entire video. To address this issue, we introduce Event-Bench, an event-oriented long video understanding benchmark built on existing datasets and human annotations. Event-Bench includes six event-related tasks and 2,190 test instances to comprehensively evaluate video event understanding ability. Additionally, we propose Video Instruction Merging~(VIM), a cost-effective method that enhances video MLLMs using merged, event-intensive video instructions, addressing the scarcity of human-annotated, event-intensive data. Extensive experiments show that the best-performing model, GPT-4o, achieves an overall accuracy of 53.33, significantly outperforming the best open-source model by 41.42%. Leveraging an effective instruction synthesis method and an adaptive model architecture, VIM surpasses both state-of-the-art open-source models and GPT-4V on the Event-Bench. All code, data, and models are publicly available at https://github.com/RUCAIBox/Event-Bench., Comment: Work on progress

Large language models (LLMs) are still struggling in aligning with human preference in complex tasks and scenarios. They are prone to overfit into the unexpected patterns or superficial styles in the training data. We conduct an empirical study that only selects the top-10\% most updated parameters in LLMs for alignment training, and see improvements in the convergence process and final performance. It indicates the existence of redundant neurons in LLMs for alignment training. To reduce its influence, we propose a low-redundant alignment method named \textbf{ALLO}, focusing on optimizing the most related neurons with the most useful supervised signals. Concretely, we first identify the neurons that are related to the human preference data by a gradient-based strategy, then identify the alignment-related key tokens by reward models for computing loss. Besides, we also decompose the alignment process into the forgetting and learning stages, where we first forget the tokens with unaligned knowledge and then learn aligned knowledge, by updating different ratios of neurons, respectively. Experimental results on 10 datasets have shown the effectiveness of ALLO. Our code and data are available at \url{https://github.com/RUCAIBox/ALLO}., Comment: 14 pages, working in progress

Recent language model-based text-to-speech (TTS) frameworks demonstrate scalability and in-context learning capabilities. However, they suffer from robustness issues due to the accumulation of errors in speech unit predictions during autoregressive language modeling. In this paper, we propose a phonetic enhanced language modeling method to improve the performance of TTS models. We leverage self-supervised representations that are phonetically rich as the training target for the autoregressive language model. Subsequently, a non-autoregressive model is employed to predict discrete acoustic codecs that contain fine-grained acoustic details. The TTS model focuses solely on linguistic modeling during autoregressive training, thereby reducing the error propagation that occurs in non-autoregressive training. Both objective and subjective evaluations validate the effectiveness of our proposed method., Comment: Accepted by Interspeech 2024

Low-Light Video Enhancement (LLVE) seeks to restore dynamic and static scenes plagued by severe invisibility and noise. One critical aspect is formulating a consistency constraint specifically for temporal-spatial illumination and appearance enhanced versions, a dimension overlooked in existing methods. In this paper, we present an innovative video Retinex-based decomposition strategy that operates without the need for explicit supervision to delineate illumination and reflectance components. We leverage dynamic cross-frame correspondences for intrinsic appearance and enforce a scene-level continuity constraint on the illumination field to yield satisfactory consistent decomposition results. To further ensure consistent decomposition, we introduce a dual-structure enhancement network featuring a novel cross-frame interaction mechanism. This mechanism can seamlessly integrate with encoder-decoder single-frame networks, incurring minimal additional parameter costs. By supervising different frames simultaneously, this network encourages them to exhibit matching decomposition features, thus achieving the desired temporal propagation. Extensive experiments are conducted on widely recognized LLVE benchmarks, covering diverse scenarios. Our framework consistently outperforms existing methods, establishing a new state-of-the-art (SOTA) performance.

We present ElastoGen, a knowledge-driven AI model that generates physically accurate 4D elastodynamics. Unlike deep models that learn from video- or image-based observations, ElastoGen leverages the principles of physics and learns from established mathematical and optimization procedures. The core idea of ElastoGen is converting the differential equation, corresponding to the nonlinear force equilibrium, into a series of iterative local convolution-like operations, which naturally fit deep architectures. We carefully build our network module following this overarching design philosophy. ElastoGen is much more lightweight in terms of both training requirements and network scale than deep generative models. Because of its alignment with actual physical procedures, ElastoGen efficiently generates accurate dynamics for a wide range of hyperelastic materials and can be easily integrated with upstream and downstream deep modules to enable end-to-end 4D generation.

We present a framework of elastic locomotion, which allows users to enliven an elastic body to produce interesting locomotion by prescribing its high-level kinematics. We formulate this problem as an inverse simulation problem and seek the optimal muscle activations to drive the body to complete the desired actions. We employ the interior-point method to model wide-area contacts between the body and the environment with logarithmic barrier penalties. The core of our framework is a mixed second-order differentiation algorithm. By combining both analytic differentiation and numerical differentiation modalities, a general-purpose second-order differentiation scheme is made possible. Specifically, we augment complex-step finite difference (CSFD) with reverse automatic differentiation (AD). We treat AD as a generic function, mapping a computing procedure to its derivative w.r.t. output loss, and promote CSFD along the AD computation. To this end, we carefully implement all the arithmetics used in elastic locomotion, from elementary functions to linear algebra and matrix operation for CSFD promotion. With this novel differentiation tool, elastic locomotion can directly exploit Newton's method and use its strong second-order convergence to find the needed activations at muscle fibers. This is not possible with existing first-order inverse or differentiable simulation techniques. We showcase a wide range of interesting locomotions of soft bodies and creatures to validate our method., Comment: 13 pages, 14 figures

Mathematical reasoning is an important capability of large language models~(LLMs) for real-world applications. To enhance this capability, existing work either collects large-scale math-related texts for pre-training, or relies on stronger LLMs (\eg GPT-4) to synthesize massive math problems. Both types of work generally lead to large costs in training or synthesis. To reduce the cost, based on open-source available texts, we propose an efficient way that trains a small LLM for math problem synthesis, to efficiently generate sufficient high-quality pre-training data. To achieve it, we create a dataset using GPT-4 to distill its data synthesis capability into the small LLM. Concretely, we craft a set of prompts based on human education stages to guide GPT-4, to synthesize problems covering diverse math knowledge and difficulty levels. Besides, we adopt the gradient-based influence estimation method to select the most valuable math-related texts. The both are fed into GPT-4 for creating the knowledge distillation dataset to train the small LLM. We leverage it to synthesize 6 million math problems for pre-training our JiuZhang3.0 model, which only needs to invoke GPT-4 API 9.3k times and pre-train on 4.6B data. Experimental results have shown that JiuZhang3.0 achieves state-of-the-art performance on several mathematical reasoning datasets, under both natural language reasoning and tool manipulation settings. Our code and data will be publicly released in \url{https://github.com/RUCAIBox/JiuZhang3.0}., Comment: 28 pages, SOTA math LLM using Well-trained Data Synthesis LLM

Large Language Models (LLMs) have shown remarkable capabilities in tasks such as summarization, arithmetic reasoning, and question answering. However, they encounter significant challenges in the domain of moral reasoning and ethical decision-making, especially in complex scenarios with multiple stakeholders. This paper introduces the Skin-in-the-Game (SKIG) framework, aimed at enhancing moral reasoning in LLMs by exploring decisions' consequences from multiple stakeholder perspectives. Central to SKIG's mechanism is simulating accountability for actions, which, alongside empathy exercises and risk assessment, is pivotal to its effectiveness. We validate SKIG's performance across various moral reasoning benchmarks with proprietary and opensource LLMs, and investigate its crucial components through extensive ablation analyses., Comment: ACL 2024, long paper

It remains a challenge to effectively control the emotion rendering in text-to-speech (TTS) synthesis. Prior studies have primarily focused on learning a global prosodic representation at the utterance level, which strongly correlates with linguistic prosody. Our goal is to construct a hierarchical emotion distribution (ED) that effectively encapsulates intensity variations of emotions at various levels of granularity, encompassing phonemes, words, and utterances. During TTS training, the hierarchical ED is extracted from the ground-truth audio and guides the predictor to establish a connection between emotional and linguistic prosody. At run-time inference, the TTS model generates emotional speech and, at the same time, provides quantitative control of emotion over the speech constituents. Both objective and subjective evaluations validate the effectiveness of the proposed framework in terms of emotion prediction and control., Comment: This is accepted to IEEE ICASSP 2024

Recent breakthroughs in large models have highlighted the critical significance of data scale, labels and modals. In this paper, we introduce MS MARCO Web Search, the first large-scale information-rich web dataset, featuring millions of real clicked query-document labels. This dataset closely mimics real-world web document and query distribution, provides rich information for various kinds of downstream tasks and encourages research in various areas, such as generic end-to-end neural indexer models, generic embedding models, and next generation information access system with large language models. MS MARCO Web Search offers a retrieval benchmark with three web retrieval challenge tasks that demand innovations in both machine learning and information retrieval system research domains. As the first dataset that meets large, real and rich data requirements, MS MARCO Web Search paves the way for future advancements in AI and system research. MS MARCO Web Search dataset is available at: https://github.com/microsoft/MS-MARCO-Web-Search., Comment: 10 pages, 6 figures, for associated dataset, see http://github.com/microsoft/MS-MARCO-Web-Search

We present X-SLAM, a real-time dense differentiable SLAM system that leverages the complex-step finite difference (CSFD) method for efficient calculation of numerical derivatives, bypassing the need for a large-scale computational graph. The key to our approach is treating the SLAM process as a differentiable function, enabling the calculation of the derivatives of important SLAM parameters through Taylor series expansion within the complex domain. Our system allows for the real-time calculation of not just the gradient, but also higher-order differentiation. This facilitates the use of high-order optimizers to achieve better accuracy and faster convergence. Building on X-SLAM, we implemented end-to-end optimization frameworks for two important tasks: camera relocalization in wide outdoor scenes and active robotic scanning in complex indoor environments. Comprehensive evaluations on public benchmarks and intricate real scenes underscore the improvements in the accuracy of camera relocalization and the efficiency of robotic navigation achieved through our task-aware optimization. The code and data are available at https://gapszju.github.io/X-SLAM., Comment: To be published in ACM SIGGRAPH 2024

Expressive voice conversion (VC) conducts speaker identity conversion for emotional speakers by jointly converting speaker identity and emotional style. Emotional style modeling for arbitrary speakers in expressive VC has not been extensively explored. Previous approaches have relied on vocoders for speech reconstruction, which makes speech quality heavily dependent on the performance of vocoders. A major challenge of expressive VC lies in emotion prosody modeling. To address these challenges, this paper proposes a fully end-to-end expressive VC framework based on a conditional denoising diffusion probabilistic model (DDPM). We utilize speech units derived from self-supervised speech models as content conditioning, along with deep features extracted from speech emotion recognition and speaker verification systems to model emotional style and speaker identity. Objective and subjective evaluations show the effectiveness of our framework. Codes and samples are publicly available., Comment: Accepted by Speaker Odyssey 2024

We present Real-time Gaussian SLAM (RTG-SLAM), a real-time 3D reconstruction system with an RGBD camera for large-scale environments using Gaussian splatting. The system features a compact Gaussian representation and a highly efficient on-the-fly Gaussian optimization scheme. We force each Gaussian to be either opaque or nearly transparent, with the opaque ones fitting the surface and dominant colors, and transparent ones fitting residual colors. By rendering depth in a different way from color rendering, we let a single opaque Gaussian well fit a local surface region without the need of multiple overlapping Gaussians, hence largely reducing the memory and computation cost. For on-the-fly Gaussian optimization, we explicitly add Gaussians for three types of pixels per frame: newly observed, with large color errors, and with large depth errors. We also categorize all Gaussians into stable and unstable ones, where the stable Gaussians are expected to well fit previously observed RGBD images and otherwise unstable. We only optimize the unstable Gaussians and only render the pixels occupied by unstable Gaussians. In this way, both the number of Gaussians to be optimized and pixels to be rendered are largely reduced, and the optimization can be done in real time. We show real-time reconstructions of a variety of large scenes. Compared with the state-of-the-art NeRF-based RGBD SLAM, our system achieves comparable high-quality reconstruction but with around twice the speed and half the memory cost, and shows superior performance in the realism of novel view synthesis and camera tracking accuracy., Comment: To be published in ACM SIGGRAPH 2024

We introduce 3D Gaussian blendshapes for modeling photorealistic head avatars. Taking a monocular video as input, we learn a base head model of neutral expression, along with a group of expression blendshapes, each of which corresponds to a basis expression in classical parametric face models. Both the neutral model and expression blendshapes are represented as 3D Gaussians, which contain a few properties to depict the avatar appearance. The avatar model of an arbitrary expression can be effectively generated by combining the neutral model and expression blendshapes through linear blending of Gaussians with the expression coefficients. High-fidelity head avatar animations can be synthesized in real time using Gaussian splatting. Compared to state-of-the-art methods, our Gaussian blendshape representation better captures high-frequency details exhibited in input video, and achieves superior rendering performance., Comment: ACM SIGGRAPH Conference Proceedings 2024

The advent of neural and Gaussian-based radiance field methods have achieved great success in the field of novel view synthesis. However, specular reflection remains non-trivial, as the high frequency radiance field is notoriously difficult to fit stably and accurately. We present a deferred shading method to effectively render specular reflection with Gaussian splatting. The key challenge comes from the environment map reflection model, which requires accurate surface normal while simultaneously bottlenecks normal estimation with discontinuous gradients. We leverage the per-pixel reflection gradients generated by deferred shading to bridge the optimization process of neighboring Gaussians, allowing nearly correct normal estimations to gradually propagate and eventually spread over all reflective objects. Our method significantly outperforms state-of-the-art techniques and concurrent work in synthesizing high-quality specular reflection effects, demonstrating a consistent improvement of peak signal-to-noise ratio (PSNR) for both synthetic and real-world scenes, while running at a frame rate almost identical to vanilla Gaussian splatting.

Undoubtedly, high-fidelity 3D hair is crucial for achieving realism, artistic expression, and immersion in computer graphics. While existing 3D hair modeling methods have achieved impressive performance, the challenge of achieving high-quality hair reconstruction persists: they either require strict capture conditions, making practical applications difficult, or heavily rely on learned prior data, obscuring fine-grained details in images. To address these challenges, we propose MonoHair,a generic framework to achieve high-fidelity hair reconstruction from a monocular video, without specific requirements for environments. Our approach bifurcates the hair modeling process into two main stages: precise exterior reconstruction and interior structure inference. The exterior is meticulously crafted using our Patch-based Multi-View Optimization (PMVO). This method strategically collects and integrates hair information from multiple views, independent of prior data, to produce a high-fidelity exterior 3D line map. This map not only captures intricate details but also facilitates the inference of the hair's inner structure. For the interior, we employ a data-driven, multi-view 3D hair reconstruction method. This method utilizes 2D structural renderings derived from the reconstructed exterior, mirroring the synthetic 2D inputs used during training. This alignment effectively bridges the domain gap between our training data and real-world data, thereby enhancing the accuracy and reliability of our interior structure inference. Lastly, we generate a strand model and resolve the directional ambiguity by our hair growth algorithm. Our experiments demonstrate that our method exhibits robustness across diverse hairstyles and achieves state-of-the-art performance. For more results, please refer to our project page https://keyuwu-cs.github.io/MonoHair/., Comment: Accepted by IEEE CVPR 2024

In this paper, we study the harmlessness alignment problem of multimodal large language models (MLLMs). We conduct a systematic empirical analysis of the harmlessness performance of representative MLLMs and reveal that the image input poses the alignment vulnerability of MLLMs. Inspired by this, we propose a novel jailbreak method named HADES, which hides and amplifies the harmfulness of the malicious intent within the text input, using meticulously crafted images. Experimental results show that HADES can effectively jailbreak existing MLLMs, which achieves an average Attack Success Rate (ASR) of 90.26% for LLaVA-1.5 and 71.60% for Gemini Pro Vision. Our code and data will be publicly released., Comment: Work in progress

Visual instruction tuning is the key to building large vision language models~(LVLMs), which can greatly improve the task generalization and solving capabilities by learning a mixture of instruction data from diverse visual tasks. Previous work mostly collects multiple existing visual instruction datasets via heuristic ways for training (even more than a million instructions), which may introduce data redundancy and enlarge the training cost. To investigate this issue, we conduct a series of empirical studies, which reveal a significant redundancy within the visual instruction datasets, and show that greatly reducing the amount of instructions from several tasks even do not affect the performance. Based on the findings, we propose a high-value data selection approach TIVE, to eliminate redundancy within the visual instruction data and reduce the training cost. In TIVE, we first estimate the instance influence score on its corresponding task, and the task difficulty score, based on the gradient-based influence functions. Then, we leverage the two kinds of scores to determine the task proportion within the selected visual instruction subset, and select high-value instances for each task, respectively. Experiments on various LVLMs show that our approach using only about 15% data can achieve comparable average performance to the full-data fine-tuned model across eight benchmarks, even surpassing it on four of the benchmarks. Our code and data will be publicly released., Comment: Under Review

A modular tensor category is a non-degenerate ribbon finite tensor category. And a ribbon factorizable Hopf algebra is exactly the Hopf algebra whose finite-dimensional representations form a modular tensor category. The goal of this paper is to construct both semisimple and non-semisimple modular categories with Hopf algebras. In particular, we study central extensions of Hopf algebras and characterize some conditions for a quotient Hopf algebra to be a factorizable one. Then we give a special way to obtain ribbon elements. As a result, we apply to some finite dimensional pointed Hopf algebras introduced by Andruskiewitsch etc and obtain two families of ribbon factorizable Hopf algebras. With little restrictions to their parameters, we prove that their representation categories are prime modular tensor categories which are not tensor equivalent to representation categories of small quantum groups. Lastly, we construct a family of semisimple ribbon factorizable Hopf algebas A(p,q), where p,q are prime numbers satisfying some conditions. And we decompose Rep(A(p,q)) into prime modular tensor categories and prove that A(p,q) can't be obtained by some obvious ways, i.e. it's not tensor product of trivial Hopf algebras (group algebras or their dual) and Drinfeld doubles.

It remains a significant challenge how to quantitatively control the expressiveness of speech emotion in speech generation. In this work, we present a novel approach for manipulating the rendering of emotions for speech generation. We propose a hierarchical emotion distribution extractor, i.e. Hierarchical ED, that quantifies the intensity of emotions at different levels of granularity. Support vector machines (SVMs) are employed to rank emotion intensity, resulting in a hierarchical emotional embedding. Hierarchical ED is subsequently integrated into the FastSpeech2 framework, guiding the model to learn emotion intensity at phoneme, word, and utterance levels. During synthesis, users can manually edit the emotional intensity of the generated voices. Both objective and subjective evaluations demonstrate the effectiveness of the proposed network in terms of fine-grained quantitative emotion editing., Comment: This is accepted to IEEE APSIPA ASC 2024

In this paper, we aim to improve the reasoning ability of large language models (LLMs) over knowledge graphs (KGs) to answer complex questions. Inspired by existing methods that design the interaction strategy between LLMs and KG, we propose an autonomous LLM-based agent framework, called KG-Agent, which enables a small LLM to actively make decisions until finishing the reasoning process over KGs. In KG-Agent, we integrate the LLM, multifunctional toolbox, KG-based executor, and knowledge memory, and develop an iteration mechanism that autonomously selects the tool then updates the memory for reasoning over KG. To guarantee the effectiveness, we leverage program language to formulate the multi-hop reasoning process over the KG, and synthesize a code-based instruction dataset to fine-tune the base LLM. Extensive experiments demonstrate that only using 10K samples for tuning LLaMA-7B can outperform state-of-the-art methods using larger LLMs or more data, on both in-domain and out-domain datasets. Our code and data will be publicly released., Comment: work in progress; efficient 7B LLM-based agent

Recent communities have seen significant progress in building photo-realistic animatable avatars from sparse multi-view videos. However, current workflows struggle to render realistic garment dynamics for loose-fitting characters as they predominantly rely on naked body models for human modeling while leaving the garment part un-modeled. This is mainly due to that the deformations yielded by loose garments are highly non-rigid, and capturing such deformations often requires dense views as supervision. In this paper, we introduce AniDress, a novel method for generating animatable human avatars in loose clothes using very sparse multi-view videos (4-8 in our setting). To allow the capturing and appearance learning of loose garments in such a situation, we employ a virtual bone-based garment rigging model obtained from physics-based simulation data. Such a model allows us to capture and render complex garment dynamics through a set of low-dimensional bone transformations. Technically, we develop a novel method for estimating temporal coherent garment dynamics from a sparse multi-view video. To build a realistic rendering for unseen garment status using coarse estimations, a pose-driven deformable neural radiance field conditioned on both body and garment motions is introduced, providing explicit control of both parts. At test time, the new garment poses can be captured from unseen situations, derived from a physics-based or neural network-based simulator to drive unseen garment dynamics. To evaluate our approach, we create a multi-view dataset that captures loose-dressed performers with diverse motions. Experiments show that our method is able to render natural garment dynamics that deviate highly from the body and generalize well to both unseen views and poses, surpassing the performance of existing methods. The code and data will be publicly available.

We demonstrate the feasibility of integrating physics-based animations of solids and fluids with 3D Gaussian Splatting (3DGS) to create novel effects in virtual scenes reconstructed using 3DGS. Leveraging the coherence of the Gaussian Splatting and Position-Based Dynamics (PBD) in the underlying representation, we manage rendering, view synthesis, and the dynamics of solids and fluids in a cohesive manner. Similar to GaussianShader, we enhance each Gaussian kernel with an added normal, aligning the kernel's orientation with the surface normal to refine the PBD simulation. This approach effectively eliminates spiky noises that arise from rotational deformation in solids. It also allows us to integrate physically based rendering to augment the dynamic surface reflections on fluids. Consequently, our framework is capable of realistically reproducing surface highlights on dynamic fluids and facilitating interactions between scene objects and fluids from new views. For more information, please visit our project page at \url{https://gaussiansplashing.github.io/}.

Reinforcement learning (RL) has been widely used in training large language models (LLMs) for preventing unexpected outputs, eg reducing harmfulness and errors. However, existing RL methods mostly adopt the instance-level reward, which is unable to provide fine-grained supervision for complex reasoning tasks, and can not focus on the few key tokens that lead to the incorrectness. To address it, we propose a new RL method named RLMEC that incorporates a generative model as the reward model, which is trained by the erroneous solution rewriting task under the minimum editing constraint, and can produce token-level rewards for RL training. Based on the generative reward model, we design the token-level RL objective for training and an imitation-based regularization for stabilizing RL process. And the both objectives focus on the learning of the key tokens for the erroneous solution, reducing the effect of other unimportant tokens. The experiment results on mathematical tasks and question-answering tasks have demonstrated the effectiveness of our approach. Our code and data are available at https://github.com/RUCAIBox/RLMEC., Comment: 18 pages, Findings of ACL2024

Recently, multi-task instruction tuning has been applied into sentence representation learning, which endows the capability of generating specific representations with the guidance of task instruction, exhibiting strong generalization ability on new tasks. However, these methods mostly neglect the potential interference problems across different tasks and instances, which may affect the training and convergence of the model. To address it, we propose a data curriculum method, namely Data-CUBE, that arranges the orders of all the multi-task data for training, to minimize the interference risks from the two views. In the task level, we aim to find the optimal task order to minimize the total cross-task interference risk, which is exactly the traveling salesman problem, hence we utilize a simulated annealing algorithm to find its solution. In the instance level, we measure the difficulty of all instances per task, then divide them into the easy-to-difficult mini-batches for training. Experiments on MTEB sentence representation evaluation tasks show that our approach can boost the performance of state-of-the-art methods. Our code and data are publicly available at the link: \url{https://github.com/RUCAIBox/Data-CUBE}., Comment: 14 pages, working in progress

Question Answering over Knowledge Graph (KGQA) aims to seek answer entities for the natural language question from a large-scale Knowledge Graph~(KG). To better perform reasoning on KG, recent work typically adopts a pre-trained language model~(PLM) to model the question, and a graph neural network~(GNN) based module to perform multi-hop reasoning on the KG. Despite the effectiveness, due to the divergence in model architecture, the PLM and GNN are not closely integrated, limiting the knowledge sharing and fine-grained feature interactions. To solve it, we aim to simplify the above two-module approach, and develop a more capable PLM that can directly support subgraph reasoning for KGQA, namely ReasoningLM. In our approach, we propose a subgraph-aware self-attention mechanism to imitate the GNN for performing structured reasoning, and also adopt an adaptation tuning strategy to adapt the model parameters with 20,000 subgraphs with synthesized questions. After adaptation, the PLM can be parameter-efficient fine-tuned on downstream tasks. Experiments show that ReasoningLM surpasses state-of-the-art models by a large margin, even with fewer updated parameters and less training data. Our codes and data are publicly available at~\url{https://github.com/RUCAIBox/ReasoningLM}., Comment: EMNLP-23-Main; simple but effective SOTA on CWQ under a weak-supervised setting

Our previously proposed MossFormer has achieved promising performance in monaural speech separation. However, it predominantly adopts a self-attention-based MossFormer module, which tends to emphasize longer-range, coarser-scale dependencies, with a deficiency in effectively modelling finer-scale recurrent patterns. In this paper, we introduce a novel hybrid model that provides the capabilities to model both long-range, coarse-scale dependencies and fine-scale recurrent patterns by integrating a recurrent module into the MossFormer framework. Instead of applying the recurrent neural networks (RNNs) that use traditional recurrent connections, we present a recurrent module based on a feedforward sequential memory network (FSMN), which is considered "RNN-free" recurrent network due to the ability to capture recurrent patterns without using recurrent connections. Our recurrent module mainly comprises an enhanced dilated FSMN block by using gated convolutional units (GCU) and dense connections. In addition, a bottleneck layer and an output layer are also added for controlling information flow. The recurrent module relies on linear projections and convolutions for seamless, parallel processing of the entire sequence. The integrated MossFormer2 hybrid model demonstrates remarkable enhancements over MossFormer and surpasses other state-of-the-art methods in WSJ0-2/3mix, Libri2Mix, and WHAM!/WHAMR! benchmarks., Comment: 5 pages, 3 figures, accepted by ICASSP 2024

Explicit pose prior models compress human poses into latent representations for using in pose-related downstream tasks. A desirable explicit pose prior model should satisfy three desirable abilities: 1) correctness, i.e. ensuring to generate physically possible poses; 2) expressiveness, i.e. ensuring to preserve details in generation; 3) controllability, meaning that generation from reference poses and explicit instructions should be convenient. Existing explicit pose prior models fail to achieve all of three properties, in special controllability. To break this situation, we propose QPoser, a highly controllable explicit pose prior model which guarantees correctness and expressiveness. In QPoser, a multi-head vector quantized autoencoder (MS-VQVAE) is proposed for obtaining expressive and distributed pose representations. Furthermore, a global-local feature integration mechanism (GLIF-AE) is utilized to disentangle the latent representation and integrate full-body information into local-joint features. Experimental results show that QPoser significantly outperforms state-of-the-art approaches in representing expressive and correct poses, meanwhile is easily to be used for detailed conditional generation from reference poses and prompting instructions.