Abstract:We present Reduced Gaussian Blendshapes Avatar (RGBAvatar), a method for reconstructing photorealistic, animatable head avatars at speeds sufficient for on-the-fly reconstruction. Unlike prior approaches that utilize linear bases from 3D morphable models (3DMM) to model Gaussian blendshapes, our method maps tracked 3DMM parameters into reduced blendshape weights with an MLP, leading to a compact set of blendshape bases. The learned compact base composition effectively captures essential facial details for specific individuals, and does not rely on the fixed base composition weights of 3DMM, leading to enhanced reconstruction quality and higher efficiency. To further expedite the reconstruction process, we develop a novel color initialization estimation method and a batch-parallel Gaussian rasterization process, achieving state-of-the-art quality with training throughput of about 630 images per second. Moreover, we propose a local-global sampling strategy that enables direct on-the-fly reconstruction, immediately reconstructing the model as video streams in real time while achieving quality comparable to offline settings. Our source code is available at https://github.com/gapszju/RGBAvatar.
Abstract:Visual generative abductive learning studies jointly training symbol-grounded neural visual generator and inducing logic rules from data, such that after learning, the visual generation process is guided by the induced logic rules. A major challenge for this task is to reduce the time cost of logic abduction during learning, an essential step when the logic symbol set is large and the logic rule to induce is complicated. To address this challenge, we propose a pre-training method for obtaining meta-rule selection policy for the recently proposed visual generative learning approach AbdGen [Peng et al., 2023], aiming at significantly reducing the candidate meta-rule set and pruning the search space. The selection model is built based on the embedding representation of both symbol grounding of cases and meta-rules, which can be effectively integrated with both neural model and logic reasoning system. The pre-training process is done on pure symbol data, not involving symbol grounding learning of raw visual inputs, making the entire learning process low-cost. An additional interesting observation is that the selection policy can rectify symbol grounding errors unseen during pre-training, which is resulted from the memorization ability of attention mechanism and the relative stability of symbolic patterns. Experimental results show that our method is able to effectively address the meta-rule selection problem for visual abduction, boosting the efficiency of visual generative abductive learning. Code is available at https://github.com/future-item/metarule-select.
Abstract:Visual instruction tuning has become the predominant technology in eliciting the multimodal task-solving capabilities of large vision-language models (LVLMs). Despite the success, as visual instructions require images as the input, it would leave the gap in inheriting the task-solving capabilities from the backbone LLMs, and make it costly to collect a large-scale dataset. To address it, we propose ViFT, a visual instruction-free fine-tuning framework for LVLMs. In ViFT, we only require the text-only instructions and image caption data during training, to separately learn the task-solving and visual perception abilities. During inference, we extract and combine the representations of the text and image inputs, for fusing the two abilities to fulfill multimodal tasks. Experimental results demonstrate that ViFT can achieve state-of-the-art performance on several visual reasoning and visual instruction following benchmarks, with rather less training data. Our code and data will be publicly released.
Abstract:The application of generative adversarial networks (GANs) has recently advanced speech super-resolution (SR) based on intermediate representations like mel-spectrograms. However, existing SR methods that typically rely on independently trained and concatenated networks may lead to inconsistent representations and poor speech quality, especially in out-of-domain scenarios. In this work, we propose HiFi-SR, a unified network that leverages end-to-end adversarial training to achieve high-fidelity speech super-resolution. Our model features a unified transformer-convolutional generator designed to seamlessly handle both the prediction of latent representations and their conversion into time-domain waveforms. The transformer network serves as a powerful encoder, converting low-resolution mel-spectrograms into latent space representations, while the convolutional network upscales these representations into high-resolution waveforms. To enhance high-frequency fidelity, we incorporate a multi-band, multi-scale time-frequency discriminator, along with a multi-scale mel-reconstruction loss in the adversarial training process. HiFi-SR is versatile, capable of upscaling any input speech signal between 4 kHz and 32 kHz to a 48 kHz sampling rate. Experimental results demonstrate that HiFi-SR significantly outperforms existing speech SR methods across both objective metrics and ABX preference tests, for both in-domain and out-of-domain scenarios (https://github.com/modelscope/ClearerVoice-Studio).
Abstract:Recently, the application of diffusion probabilistic models has advanced speech enhancement through generative approaches. However, existing diffusion-based methods have focused on the generation process in high-dimensional waveform or spectral domains, leading to increased generation complexity and slower inference speeds. Additionally, these methods have primarily modelled clean speech distributions, with limited exploration of noise distributions, thereby constraining the discriminative capability of diffusion models for speech enhancement. To address these issues, we propose a novel approach that integrates a conditional latent diffusion model (cLDM) with dual-context learning (DCL). Our method utilizes a variational autoencoder (VAE) to compress mel-spectrograms into a low-dimensional latent space. We then apply cLDM to transform the latent representations of both clean speech and background noise into Gaussian noise by the DCL process, and a parameterized model is trained to reverse this process, conditioned on noisy latent representations and text embeddings. By operating in a lower-dimensional space, the latent representations reduce the complexity of the generation process, while the DCL process enhances the model's ability to handle diverse and unseen noise environments. Our experiments demonstrate the strong performance of the proposed approach compared to existing diffusion-based methods, even with fewer iterative steps, and highlight the superior generalization capability of our models to out-of-domain noise datasets (https://github.com/modelscope/ClearerVoice-Studio).
Abstract:Effective pre-training of large language models (LLMs) has been challenging due to the immense resource demands and the complexity of the technical processes involved. This paper presents a detailed technical report on YuLan-Mini, a highly capable base model with 2.42B parameters that achieves top-tier performance among models of similar parameter scale. Our pre-training approach focuses on enhancing training efficacy through three key technical contributions: an elaborate data pipeline combines data cleaning with data schedule strategies, a robust optimization method to mitigate training instability, and an effective annealing approach that incorporates targeted data selection and long context training. Remarkably, YuLan-Mini, trained on 1.08T tokens, achieves performance comparable to industry-leading models that require significantly more data. To facilitate reproduction, we release the full details of the data composition for each training phase. Project details can be accessed at the following link: https://github.com/RUC-GSAI/YuLan-Mini.
Abstract:Emotional text-to-speech synthesis (TTS) aims to generate realistic emotional speech from input text. However, quantitatively controlling multi-level emotion rendering remains challenging. In this paper, we propose a diffusion-based emotional TTS framework with a novel approach for emotion intensity modeling to facilitate fine-grained control over emotion rendering at the phoneme, word, and utterance levels. We introduce a hierarchical emotion distribution (ED) extractor that captures a quantifiable ED embedding across different speech segment levels. Additionally, we explore various acoustic features and assess their impact on emotion intensity modeling. During TTS training, the hierarchical ED embedding effectively captures the variance in emotion intensity from the reference audio and correlates it with linguistic and speaker information. The TTS model not only generates emotional speech during inference, but also quantitatively controls the emotion rendering over the speech constituents. Both objective and subjective evaluations demonstrate the effectiveness of our framework in terms of speech quality, emotional expressiveness, and hierarchical emotion control.
Abstract:The real estate market relies heavily on structured data, such as property details, market trends, and price fluctuations. However, the lack of specialized Tabular Question Answering datasets in this domain limits the development of automated question-answering systems. To fill this gap, we introduce RETQA, the first large-scale open-domain Chinese Tabular Question Answering dataset for Real Estate. RETQA comprises 4,932 tables and 20,762 question-answer pairs across 16 sub-fields within three major domains: property information, real estate company finance information and land auction information. Compared with existing tabular question answering datasets, RETQA poses greater challenges due to three key factors: long-table structures, open-domain retrieval, and multi-domain queries. To tackle these challenges, we propose the SLUTQA framework, which integrates large language models with spoken language understanding tasks to enhance retrieval and answering accuracy. Extensive experiments demonstrate that SLUTQA significantly improves the performance of large language models on RETQA by in-context learning. RETQA and SLUTQA provide essential resources for advancing tabular question answering research in the real estate domain, addressing critical challenges in open-domain and long-table question-answering. The dataset and code are publicly available at \url{https://github.com/jensen-w/RETQA}.
Abstract:High-quality material generation is key for virtual environment authoring and inverse rendering. We propose MaterialPicker, a multi-modal material generator leveraging a Diffusion Transformer (DiT) architecture, improving and simplifying the creation of high-quality materials from text prompts and/or photographs. Our method can generate a material based on an image crop of a material sample, even if the captured surface is distorted, viewed at an angle or partially occluded, as is often the case in photographs of natural scenes. We further allow the user to specify a text prompt to provide additional guidance for the generation. We finetune a pre-trained DiT-based video generator into a material generator, where each material map is treated as a frame in a video sequence. We evaluate our approach both quantitatively and qualitatively and show that it enables more diverse material generation and better distortion correction than previous work.
Abstract:There have been recent efforts to extend the Chain-of-Thought (CoT) paradigm to Multimodal Large Language Models (MLLMs) by finding visual clues in the input scene, advancing the visual reasoning ability of MLLMs. However, current approaches are specially designed for the tasks where clue finding plays a major role in the whole reasoning process, leading to the difficulty in handling complex visual scenes where clue finding does not actually simplify the whole reasoning task. To deal with this challenge, we propose a new visual reasoning paradigm enabling MLLMs to autonomously modify the input scene to new ones based on its reasoning status, such that CoT is reformulated as conducting simple closed-loop decision-making and reasoning steps under a sequence of imagined visual scenes, leading to natural and general CoT construction. To implement this paradigm, we introduce a novel plug-and-play imagination space, where MLLMs conduct visual modifications through operations like focus, ignore, and transform based on their native reasoning ability without specific training. We validate our approach through a benchmark spanning dense counting, simple jigsaw puzzle solving, and object placement, challenging the reasoning ability beyond clue finding. The results verify that while existing techniques fall short, our approach enables MLLMs to effectively reason step by step through autonomous imagination. Project page: https://future-item.github.io/autoimagine-site.