Zhongyuan Wang

Zhuo Wang, Zezheng Wang, Zitong Yu et al.

With diverse presentation attacks emerging continually, generalizable face anti-spoofing (FAS) has drawn growing attention. Most existing methods implement domain generalization (DG) on the complete representations. However, different image statistics may have unique properties for the FAS tasks. In this work, we separate the complete representation into content and style ones. A novel Shuffled Style Assembly Network (SSAN) is proposed to extract and reassemble different content and style features for a stylized feature space. Then, to obtain a generalized representation, a contrastive learning strategy is developed to emphasize liveness-related style information while suppress the domain-specific one. Finally, the representations of the correct assemblies are used to distinguish between living and spoofing during the inferring. On the other hand, despite the decent performance, there still exists a gap between academia and industry, due to the difference in data quantity and distribution. Thus, a new large-scale benchmark for FAS is built up to further evaluate the performance of algorithms in reality. Both qualitative and quantitative results on existing and proposed benchmarks demonstrate the effectiveness of our methods. The codes will be available at https://github.com/wangzhuo2019/SSAN.

Weikai Yang, Xi Ye, Xingxing Zhang et al. · microsoft-research

The base learners and labeled samples (shots) in an ensemble few-shot classifier greatly affect the model performance. When the performance is not satisfactory, it is usually difficult to understand the underlying causes and make improvements. To tackle this issue, we propose a visual analysis method, FSLDiagnotor. Given a set of base learners and a collection of samples with a few shots, we consider two problems: 1) finding a subset of base learners that well predict the sample collections; and 2) replacing the low-quality shots with more representative ones to adequately represent the sample collections. We formulate both problems as sparse subset selection and develop two selection algorithms to recommend appropriate learners and shots, respectively. A matrix visualization and a scatterplot are combined to explain the recommended learners and shots in context and facilitate users in adjusting them. Based on the adjustment, the algorithm updates the recommendation results for another round of improvement. Two case studies are conducted to demonstrate that FSLDiagnotor helps build a few-shot classifier efficiently and increases the accuracy by 12% and 21%, respectively.

Kui Jiang, Zhongyuan Wang, Chen Chen et al.

Convolutional neural network (CNN) and Transformer have achieved great success in multimedia applications. However, little effort has been made to effectively and efficiently harmonize these two architectures to satisfy image deraining. This paper aims to unify these two architectures to take advantage of their learning merits for image deraining. In particular, the local connectivity and translation equivariance of CNN and the global aggregation ability of self-attention (SA) in Transformer are fully exploited for specific local context and global structure representations. Based on the observation that rain distribution reveals the degradation location and degree, we introduce degradation prior to help background recovery and accordingly present the association refinement deraining scheme. A novel multi-input attention module (MAM) is proposed to associate rain perturbation removal and background recovery. Moreover, we equip our model with effective depth-wise separable convolutions to learn the specific feature representations and trade off computational complexity. Extensive experiments show that our proposed method (dubbed as ELF) outperforms the state-of-the-art approach (MPRNet) by 0.25 dB on average, but only accounts for 11.7\% and 42.1\% of its computational cost and parameters. The source code is available at https://github.com/kuijiang94/Magic-ELF.

Wejia Wu, Zhuang Li, Jiahong Li et al.

Video text spotting(VTS) is the task that requires simultaneously detecting, tracking and recognizing text in the video. Existing video text spotting methods typically develop sophisticated pipelines and multiple models, which is not friend for real-time applications. Here we propose a real-time end-to-end video text spotter with Contrastive Representation learning (CoText). Our contributions are three-fold: 1) CoText simultaneously address the three tasks (e.g., text detection, tracking, recognition) in a real-time end-to-end trainable framework. 2) With contrastive learning, CoText models long-range dependencies and learning temporal information across multiple frames. 3) A simple, lightweight architecture is designed for effective and accurate performance, including GPU-parallel detection post-processing, CTC-based recognition head with Masked RoI. Extensive experiments show the superiority of our method. Especially, CoText achieves an video text spotting IDF1 of 72.0% at 41.0 FPS on ICDAR2015video, with 10.5% and 32.0 FPS improvement the previous best method. The code can be found at github.com/weijiawu/CoText.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large Language Models (LLMs) represent a significant stride toward Artificial General Intelligence. As scaling laws underscore the potential of increasing model sizes, the academic community has intensified its investigations into LLMs with capacities exceeding 50 billion parameters. This technical report builds on our prior work with Tele-FLM (also known as FLM-2), a publicly available 52-billion-parameter model. We delve into two primary areas: we first discuss our observation of Supervised Fine-tuning (SFT) on Tele-FLM-52B, which supports the "less is more" approach for SFT data construction; second, we demonstrate our experiments and analyses on the best practices for progressively growing a model from 52 billion to 102 billion, and subsequently to 1 trillion parameters. We will open-source a 1T model checkpoint, namely Tele-FLM-1T, to advance further training and research.

Xing Wu, Chaochen Gao, Zijia Lin et al.

Contrastive learning has been extensively studied in sentence embedding learning, which assumes that the embeddings of different views of the same sentence are closer. The constraint brought by this assumption is weak, and a good sentence representation should also be able to reconstruct the original sentence fragments. Therefore, this paper proposes an information-aggregated contrastive learning framework for learning unsupervised sentence embeddings, termed InfoCSE. InfoCSE forces the representation of [CLS] positions to aggregate denser sentence information by introducing an additional Masked language model task and a well-designed network. We evaluate the proposed InfoCSE on several benchmark datasets w.r.t the semantic text similarity (STS) task. Experimental results show that InfoCSE outperforms SimCSE by an average Spearman correlation of 2.60% on BERT-base, and 1.77% on BERT-large, achieving state-of-the-art results among unsupervised sentence representation learning methods. Our code are available at https://github.com/caskcsg/sentemb/tree/main/InfoCSE.

Xing Wu, Guangyuan Ma, Meng Lin et al.

Dense passage retrieval aims to retrieve the relevant passages of a query from a large corpus based on dense representations (i.e., vectors) of the query and the passages. Recent studies have explored improving pre-trained language models to boost dense retrieval performance. This paper proposes CoT-MAE (ConTextual Masked Auto-Encoder), a simple yet effective generative pre-training method for dense passage retrieval. CoT-MAE employs an asymmetric encoder-decoder architecture that learns to compress the sentence semantics into a dense vector through self-supervised and context-supervised masked auto-encoding. Precisely, self-supervised masked auto-encoding learns to model the semantics of the tokens inside a text span, and context-supervised masked auto-encoding learns to model the semantical correlation between the text spans. We conduct experiments on large-scale passage retrieval benchmarks and show considerable improvements over strong baselines, demonstrating the high efficiency of CoT-MAE. Our code is available at https://github.com/caskcsg/ir/tree/main/cotmae.

Xing Wu, Chaochen Gao, Zijia Lin et al.

Video language pre-training methods have mainly adopted sparse sampling techniques to alleviate the temporal redundancy of videos. Though effective, sparse sampling still suffers inter-modal redundancy: visual redundancy and textual redundancy. Compared with highly generalized text, sparsely sampled frames usually contain text-independent portions, called visual redundancy. Sparse sampling is also likely to miss important frames corresponding to some text portions, resulting in textual redundancy. Inter-modal redundancy leads to a mismatch of video and text information, hindering the model from better learning the shared semantics across modalities. To alleviate it, we propose Redundancy-aware Video-language Pre-training. We design a redundancy measurement of video patches and text tokens by calculating the cross-modal minimum dis-similarity. Then, we penalize the highredundant video patches and text tokens through a proposed redundancy-aware contrastive learning. We evaluate our method on four benchmark datasets, MSRVTT, MSVD, DiDeMo, and LSMDC, achieving a significant improvement over the previous stateof-the-art results. Our code are available at https://github.com/caskcsg/VLP/tree/main/RaP.

Zhijie Nie, Richong Zhang, Zhongyuan Wang et al.

Current methods for Knowledge-Based Question Answering (KBQA) usually rely on complex training techniques and model frameworks, leading to many limitations in practical applications. Recently, the emergence of In-Context Learning (ICL) capabilities in Large Language Models (LLMs) provides a simple and training-free semantic parsing paradigm for KBQA: Given a small number of questions and their labeled logical forms as demo examples, LLMs can understand the task intent and generate the logic form for a new question. However, current powerful LLMs have little exposure to logic forms during pre-training, resulting in a high format error rate. To solve this problem, we propose a code-style in-context learning method for KBQA, which converts the generation process of unfamiliar logical form into the more familiar code generation process for LLMs. Experimental results on three mainstream datasets show that our method dramatically mitigated the formatting error problem in generating logic forms while realizing a new SOTA on WebQSP, GrailQA, and GraphQ under the few-shot setting. The code and supplementary files are released at https://github.com/Arthurizijar/KB-Coder .

Taiping Qu, Hongkai Zhang, Lantian Zhang et al.

Cardiac magnetic resonance (CMR) is a cornerstone for diagnosing cardiovascular disease. However, it remains underutilized due to complex, time-consuming interpretation across multi-sequences, phases, quantitative measures that heavily reliant on specialized expertise. Here, we present BAAI Cardiac Agent, a multimodal intelligent system designed for end-to-end CMR interpretation. The agent integrates specialized cardiac expert models to perform automated segmentation of cardiac structures, functional quantification, tissue characterization and disease diagnosis, and generates structured clinical reports within a unified workflow. Evaluated on CMR datasets from two hospitals (2413 patients) spanning 7-types of major cardiovascular diseases, the agent achieved an area under the receiver-operating-characteristic curve exceeding 0.93 internally and 0.81 externally. In the task of estimating left ventricular function indices, the results generated by this system for core parameters such as ejection fraction, stroke volume, and left ventricular mass are highly consistent with clinical reports, with Pearson correlation coefficients all exceeding 0.90. The agent outperformed state-of-the-art models in segmentation and diagnostic tasks, and generated clinical reports showing high concordance with expert radiologists (six readers across three experience levels). By dynamically orchestrating expert models for coordinated multimodal analysis, this agent framework enables accurate, efficient CMR interpretation and highlights its potentials for complex clinical imaging workflows. Code is available at https://github.com/plantain-herb/Cardiac-Agent.

Qi Fan, Xin Tao, Lei Ke et al.

The Segment Anything Model (SAM) achieves remarkable promptable segmentation given high-quality prompts which, however, often require good skills to specify. To make SAM robust to casual prompts, this paper presents the first comprehensive analysis on SAM's segmentation stability across a diverse spectrum of prompt qualities, notably imprecise bounding boxes and insufficient points. Our key finding reveals that given such low-quality prompts, SAM's mask decoder tends to activate image features that are biased towards the background or confined to specific object parts. To mitigate this issue, our key idea consists of calibrating solely SAM's mask attention by adjusting the sampling locations and amplitudes of image features, while the original SAM model architecture and weights remain unchanged. Consequently, our deformable sampling plugin (DSP) enables SAM to adaptively shift attention to the prompted target regions in a data-driven manner, facilitated by our effective robust training strategy (RTS). During inference, dynamic routing plugin (DRP) is proposed that toggles SAM between the deformable and regular grid sampling modes, conditioned on the input prompt quality. Thus, our solution, termed Stable-SAM, offers several advantages: 1) improved SAM's segmentation stability across a wide range of prompt qualities, while 2) retaining SAM's powerful promptable segmentation efficiency and generality, with 3) minimal learnable parameters (0.08 M) and fast adaptation (by 1 training epoch). Extensive experiments across multiple datasets validate the effectiveness and advantages of our approach, underscoring Stable-SAM as a more robust solution for segmenting anything. Codes will be released upon acceptance. https://github.com/fanq15/Stable-SAM

Lingfeng Zhang, Xiaoshuai Hao, Qinwen Xu et al.

Vision-and-language navigation (VLN) is a key task in Embodied AI, requiring agents to navigate diverse and unseen environments while following natural language instructions. Traditional approaches rely heavily on historical observations as spatio-temporal contexts for decision making, leading to significant storage and computational overhead. In this paper, we introduce MapNav, a novel end-to-end VLN model that leverages Annotated Semantic Map (ASM) to replace historical frames. Specifically, our approach constructs a top-down semantic map at the start of each episode and update it at each timestep, allowing for precise object mapping and structured navigation information. Then, we enhance this map with explicit textual labels for key regions, transforming abstract semantics into clear navigation cues and generate our ASM. MapNav agent using the constructed ASM as input, and use the powerful end-to-end capabilities of VLM to empower VLN. Extensive experiments demonstrate that MapNav achieves state-of-the-art (SOTA) performance in both simulated and real-world environments, validating the effectiveness of our method. Moreover, we will release our ASM generation source code and dataset to ensure reproducibility, contributing valuable resources to the field. We believe that our proposed MapNav can be used as a new memory representation method in VLN, paving the way for future research in this field.

Haojie Pan, Zepeng Zhai, Yuzhou Zhang et al.

Online encyclopedias, such as Wikipedia, have been well-developed and researched in the last two decades. One can find any attributes or other information of a wiki item on a wiki page edited by a community of volunteers. However, the traditional text, images and tables can hardly express some aspects of an wiki item. For example, when we talk about ``Shiba Inu'', one may care more about ``How to feed it'' or ``How to train it not to protect its food''. Currently, short-video platforms have become a hallmark in the online world. Whether you're on TikTok, Instagram, Kuaishou, or YouTube Shorts, short-video apps have changed how we consume and create content today. Except for producing short videos for entertainment, we can find more and more authors sharing insightful knowledge widely across all walks of life. These short videos, which we call knowledge videos, can easily express any aspects (e.g. hair or how-to-feed) consumers want to know about an item (e.g. Shiba Inu), and they can be systematically analyzed and organized like an online encyclopedia. In this paper, we propose Kuaipedia, a large-scale multi-modal encyclopedia consisting of items, aspects, and short videos lined to them, which was extracted from billions of videos of Kuaishou (Kwai), a well-known short-video platform in China. We first collected items from multiple sources and mined user-centered aspects from millions of users' queries to build an item-aspect tree. Then we propose a new task called ``multi-modal item-aspect linking'' as an expansion of ``entity linking'' to link short videos into item-aspect pairs and build the whole short-video encyclopedia. Intrinsic evaluations show that our encyclopedia is of large scale and highly accurate. We also conduct sufficient extrinsic experiments to show how Kuaipedia can help fundamental applications such as entity typing and entity linking.

Jiaxin Deng, Dong Shen, Haojie Pan et al.

Video understanding is an important task in short video business platforms and it has a wide application in video recommendation and classification. Most of the existing video understanding works only focus on the information that appeared within the video content, including the video frames, audio and text. However, introducing common sense knowledge from the external Knowledge Graph (KG) dataset is essential for video understanding when referring to the content which is less relevant to the video. Owing to the lack of video knowledge graph dataset, the work which integrates video understanding and KG is rare. In this paper, we propose a heterogeneous dataset that contains the multi-modal video entity and fruitful common sense relations. This dataset also provides multiple novel video inference tasks like the Video-Relation-Tag (VRT) and Video-Relation-Video (VRV) tasks. Furthermore, based on this dataset, we propose an end-to-end model that jointly optimizes the video understanding objective with knowledge graph embedding, which can not only better inject factual knowledge into video understanding but also generate effective multi-modal entity embedding for KG. Comprehensive experiments indicate that combining video understanding embedding with factual knowledge benefits the content-based video retrieval performance. Moreover, it also helps the model generate better knowledge graph embedding which outperforms traditional KGE-based methods on VRT and VRV tasks with at least 42.36% and 17.73% improvement in HITS@10.

Haotian Yang, Mingwu Zheng, Wanquan Feng et al.

In this paper, we propose a novel framework, Tracking-free Relightable Avatar (TRAvatar), for capturing and reconstructing high-fidelity 3D avatars. Compared to previous methods, TRAvatar works in a more practical and efficient setting. Specifically, TRAvatar is trained with dynamic image sequences captured in a Light Stage under varying lighting conditions, enabling realistic relighting and real-time animation for avatars in diverse scenes. Additionally, TRAvatar allows for tracking-free avatar capture and obviates the need for accurate surface tracking under varying illumination conditions. Our contributions are two-fold: First, we propose a novel network architecture that explicitly builds on and ensures the satisfaction of the linear nature of lighting. Trained on simple group light captures, TRAvatar can predict the appearance in real-time with a single forward pass, achieving high-quality relighting effects under illuminations of arbitrary environment maps. Second, we jointly optimize the facial geometry and relightable appearance from scratch based on image sequences, where the tracking is implicitly learned. This tracking-free approach brings robustness for establishing temporal correspondences between frames under different lighting conditions. Extensive qualitative and quantitative experiments demonstrate that our framework achieves superior performance for photorealistic avatar animation and relighting.

Chunyu Qiang, Peng Yang, Hao Che et al.

Cross-speaker style transfer in speech synthesis aims at transferring a style from source speaker to synthesised speech of a target speaker's timbre. Most previous approaches rely on data with style labels, but manually-annotated labels are expensive and not always reliable. In response to this problem, we propose Style-Label-Free, a cross-speaker style transfer method, which can realize the style transfer from source speaker to target speaker without style labels. Firstly, a reference encoder structure based on quantized variational autoencoder (Q-VAE) and style bottleneck is designed to extract discrete style representations. Secondly, a speaker-wise batch normalization layer is proposed to reduce the source speaker leakage. In order to improve the style extraction ability of the reference encoder, a style invariant and contrastive data augmentation method is proposed. Experimental results show that the method outperforms the baseline. We provide a website with audio samples.

Xiaohan Zou, Changqiao Wu, Lele Cheng et al.

Most existing methods in vision-language retrieval match two modalities by either comparing their global feature vectors which misses sufficient information and lacks interpretability, detecting objects in images or videos and aligning the text with fine-grained features which relies on complicated model designs, or modeling fine-grained interaction via cross-attention upon visual and textual tokens which suffers from inferior efficiency. To address these limitations, some recent works simply aggregate the token-wise similarities to achieve fine-grained alignment, but they lack intuitive explanations as well as neglect the relationships between token-level features and global representations with high-level semantics. In this work, we rethink fine-grained cross-modal alignment and devise a new model-agnostic formulation for it. We additionally demystify the recent popular works and subsume them into our scheme. Furthermore, inspired by optimal transport theory, we introduce TokenFlow, an instantiation of the proposed scheme. By modifying only the similarity function, the performance of our method is comparable to the SoTA algorithms with heavy model designs on major video-text retrieval benchmarks. The visualization further indicates that TokenFlow successfully leverages the fine-grained information and achieves better interpretability.

Jiaxin Ai, Zhongyuan Wang, Baojin Huang et al.

Deepfake face not only violates the privacy of personal identity, but also confuses the public and causes huge social harm. The current deepfake detection only stays at the level of distinguishing true and false, and cannot trace the original genuine face corresponding to the fake face, that is, it does not have the ability to trace the source of evidence. The deepfake countermeasure technology for judicial forensics urgently calls for deepfake traceability. This paper pioneers an interesting question about face deepfake, active forensics that "know it and how it happened". Given that deepfake faces do not completely discard the features of original faces, especially facial expressions and poses, we argue that original faces can be approximately speculated from their deepfake counterparts. Correspondingly, we design a disentangling reversing network that decouples latent space features of deepfake faces under the supervision of fake-original face pair samples to infer original faces in reverse.

Chunyu Qiang, Peng Yang, Hao Che et al.

Conversion of Chinese Grapheme-to-Phoneme (G2P) plays an important role in Mandarin Chinese Text-To-Speech (TTS) systems, where one of the biggest challenges is the task of polyphone disambiguation. Most of the previous polyphone disambiguation models are trained on manually annotated datasets, and publicly available datasets for polyphone disambiguation are scarce. In this paper we propose a simple back-translation-style data augmentation method for mandarin Chinese polyphone disambiguation, utilizing a large amount of unlabeled text data. Inspired by the back-translation technique proposed in the field of machine translation, we build a Grapheme-to-Phoneme (G2P) model to predict the pronunciation of polyphonic character, and a Phoneme-to-Grapheme (P2G) model to predict pronunciation into text. Meanwhile, a window-based matching strategy and a multi-model scoring strategy are proposed to judge the correctness of the pseudo-label. We design a data balance strategy to improve the accuracy of some typical polyphonic characters in the training set with imbalanced distribution or data scarcity. The experimental result shows the effectiveness of the proposed back-translation-style data augmentation method.

Wanfeng Zheng, Qiang Li, Xiaoyan Guo et al.

Text-driven image manipulation is developed since the vision-language model (CLIP) has been proposed. Previous work has adopted CLIP to design a text-image consistency-based objective to address this issue. However, these methods require either test-time optimization or image feature cluster analysis for single-mode manipulation direction. In this paper, we manage to achieve inference-time optimization-free diverse manipulation direction mining by bridging CLIP and StyleGAN through Latent Alignment (CSLA). More specifically, our efforts consist of three parts: 1) a data-free training strategy to train latent mappers to bridge the latent space of CLIP and StyleGAN; 2) for more precise mapping, temporal relative consistency is proposed to address the knowledge distribution bias problem among different latent spaces; 3) to refine the mapped latent in s space, adaptive style mixing is also proposed. With this mapping scheme, we can achieve GAN inversion, text-to-image generation and text-driven image manipulation. Qualitative and quantitative comparisons are made to demonstrate the effectiveness of our method.

Jing Fang, Yinbo Yu, Zhongyuan Wang et al.

Image super-resolution (SR) is a technique to recover lost high-frequency information in low-resolution (LR) images. Spatial-domain information has been widely exploited to implement image SR, so a new trend is to involve frequency-domain information in SR tasks. Besides, image SR is typically application-oriented and various computer vision tasks call for image arbitrary magnification. Therefore, in this paper, we study image features in the frequency domain to design a novel scale-arbitrary image SR network. First, we statistically analyze LR-HR image pairs of several datasets under different scale factors and find that the high-frequency spectra of different images under different scale factors suffer from different degrees of degradation, but the valid low-frequency spectra tend to be retained within a certain distribution range. Then, based on this finding, we devise an adaptive scale-aware feature division mechanism using deep reinforcement learning, which can accurately and adaptively divide the frequency spectrum into the low-frequency part to be retained and the high-frequency one to be recovered. Finally, we design a scale-aware feature recovery module to capture and fuse multi-level features for reconstructing the high-frequency spectrum at arbitrary scale factors. Extensive experiments on public datasets show the superiority of our method compared with state-of-the-art methods.

Lei Lin, Jiayi Fu, Pengli Liu et al.

Although chain-of-thought (CoT) prompting combined with language models has achieved encouraging results on complex reasoning tasks, the naive greedy decoding used in CoT prompting usually causes the repetitiveness and local optimality. To address this shortcoming, ensemble-optimization tries to obtain multiple reasoning paths to get the final answer assembly. However, current ensemble-optimization methods either simply employ rule-based post-processing such as \textit{self-consistency}, or train an additional model based on several task-related human annotations to select the best one among multiple reasoning paths, yet fail to generalize to realistic settings where the type of input questions is unknown or the answer format of reasoning paths is unknown. To avoid their limitations, we propose \textbf{Self-Agreement}, a generalizable ensemble-optimization method applying in almost all scenarios where the type of input questions and the answer format of reasoning paths may be known or unknown. Self-agreement firstly samples from language model's decoder to generate a \textit{diverse} set of reasoning paths, and subsequently prompts the language model \textit{one more time} to determine the optimal answer by selecting the most \textit{agreed} answer among the sampled reasoning paths. Self-agreement simultaneously achieves remarkable performance on six public reasoning benchmarks and superior generalization capabilities.

Zhipeng Chen, Yingqian Min, Beichen Zhang et al.

In this report, we present the third technical report on the development of slow-thinking models as part of the STILL project. As the technical pathway becomes clearer, scaling RL training has become a central technique for implementing such reasoning models. We systematically experiment with and document the effects of various factors influencing RL training, conducting experiments on both base models and fine-tuned models. Specifically, we demonstrate that our RL training approach consistently improves the Qwen2.5-32B base models, enhancing both response length and test accuracy. Furthermore, we show that even when a model like DeepSeek-R1-Distill-Qwen-1.5B has already achieved a high performance level, it can be further refined through RL training, reaching an accuracy of 39.33% on AIME 2024. Beyond RL training, we also explore the use of tool manipulation, finding that it significantly boosts the reasoning performance of large reasoning models. This approach achieves a remarkable accuracy of 86.67% with greedy search on AIME 2024, underscoring its effectiveness in enhancing model capabilities. We release our resources at the STILL project website: https://github.com/RUCAIBox/Slow_Thinking_with_LLMs.

Chunyu Qiang, Peng Yang, Hao Che et al.

Cross-speaker style transfer in speech synthesis aims at transferring a style from source speaker to synthesized speech of a target speaker's timbre. In most previous methods, the synthesized fine-grained prosody features often represent the source speaker's average style, similar to the one-to-many problem(i.e., multiple prosody variations correspond to the same text). In response to this problem, a strength-controlled semi-supervised style extractor is proposed to disentangle the style from content and timbre, improving the representation and interpretability of the global style embedding, which can alleviate the one-to-many mapping and data imbalance problems in prosody prediction. A hierarchical prosody predictor is proposed to improve prosody modeling. We find that better style transfer can be achieved by using the source speaker's prosody features that are easily predicted. Additionally, a speaker-transfer-wise cycle consistency loss is proposed to assist the model in learning unseen style-timbre combinations during the training phase. Experimental results show that the method outperforms the baseline. We provide a website with audio samples.

Jikang Cheng, Ying Zhang, Qin Zou et al.

Learning intrinsic bias from limited data has been considered the main reason for the failure of deepfake detection with generalizability. Apart from the discovered content and specific-forgery bias, we reveal a novel spatial bias, where detectors inertly anticipate observing structural forgery clues appearing at the image center, also can lead to the poor generalization of existing methods. We present ED$^4$, a simple and effective strategy, to address aforementioned biases explicitly at the data level in a unified framework rather than implicit disentanglement via network design. In particular, we develop ClockMix to produce facial structure preserved mixtures with arbitrary samples, which allows the detector to learn from an exponentially extended data distribution with much more diverse identities, backgrounds, local manipulation traces, and the co-occurrence of multiple forgery artifacts. We further propose the Adversarial Spatial Consistency Module (AdvSCM) to prevent extracting features with spatial bias, which adversarially generates spatial-inconsistent images and constrains their extracted feature to be consistent. As a model-agnostic debiasing strategy, ED$^4$ is plug-and-play: it can be integrated with various deepfake detectors to obtain significant benefits. We conduct extensive experiments to demonstrate its effectiveness and superiority over existing deepfake detection approaches.

Jiayi Fu, Lei Lin, Xiaoyang Gao et al.

Recent advancements in large language models (LLMs) have demonstrated remarkable abilities in handling a variety of natural language processing (NLP) downstream tasks, even on mathematical tasks requiring multi-step reasoning. In this report, we introduce the KwaiYiiMath which enhances the mathematical reasoning abilities of KwaiYiiBase1, by applying Supervised Fine-Tuning (SFT) and Reinforced Learning from Human Feedback (RLHF), including on both English and Chinese mathematical tasks. Meanwhile, we also constructed a small-scale Chinese primary school mathematics test set (named KMath), consisting of 188 examples to evaluate the correctness of the problem-solving process generated by the models. Empirical studies demonstrate that KwaiYiiMath can achieve state-of-the-art (SOTA) performance on GSM8k, CMath, and KMath compared with the similar size models, respectively.

Tao Zhang, Xingye Tian, Yikang Zhou et al.

We present the \textbf{D}ecoupled \textbf{VI}deo \textbf{S}egmentation (DVIS) framework, a novel approach for the challenging task of universal video segmentation, including video instance segmentation (VIS), video semantic segmentation (VSS), and video panoptic segmentation (VPS). Unlike previous methods that model video segmentation in an end-to-end manner, our approach decouples video segmentation into three cascaded sub-tasks: segmentation, tracking, and refinement. This decoupling design allows for simpler and more effective modeling of the spatio-temporal representations of objects, especially in complex scenes and long videos. Accordingly, we introduce two novel components: the referring tracker and the temporal refiner. These components track objects frame by frame and model spatio-temporal representations based on pre-aligned features. To improve the tracking capability of DVIS, we propose a denoising training strategy and introduce contrastive learning, resulting in a more robust framework named DVIS++. Furthermore, we evaluate DVIS++ in various settings, including open vocabulary and using a frozen pre-trained backbone. By integrating CLIP with DVIS++, we present OV-DVIS++, the first open-vocabulary universal video segmentation framework. We conduct extensive experiments on six mainstream benchmarks, including the VIS, VSS, and VPS datasets. Using a unified architecture, DVIS++ significantly outperforms state-of-the-art specialized methods on these benchmarks in both close- and open-vocabulary settings. Code:~\url{https://github.com/zhang-tao-whu/DVIS_Plus}.

Jikang Cheng, Zhiyuan Yan, Ying Zhang et al.

The generalization ability of deepfake detectors is vital for their applications in real-world scenarios. One effective solution to enhance this ability is to train the models with manually-blended data, which we termed "blendfake", encouraging models to learn generic forgery artifacts like blending boundary. Interestingly, current SoTA methods utilize blendfake without incorporating any deepfake data in their training process. This is likely because previous empirical observations suggest that vanilla hybrid training (VHT), which combines deepfake and blendfake data, results in inferior performance to methods using only blendfake data (so-called "1+1<2"). Therefore, a critical question arises: Can we leave deepfake behind and rely solely on blendfake data to train an effective deepfake detector? Intuitively, as deepfakes also contain additional informative forgery clues (e.g., deep generative artifacts), excluding all deepfake data in training deepfake detectors seems counter-intuitive. In this paper, we rethink the role of blendfake in detecting deepfakes and formulate the process from "real to blendfake to deepfake" to be a progressive transition. Specifically, blendfake and deepfake can be explicitly delineated as the oriented pivot anchors between "real-to-fake" transitions. The accumulation of forgery information should be oriented and progressively increasing during this transition process. To this end, we propose an Oriented Progressive Regularizor (OPR) to establish the constraints that compel the distribution of anchors to be discretely arranged. Furthermore, we introduce feature bridging to facilitate the smooth transition between adjacent anchors. Extensive experiments confirm that our design allows leveraging forgery information from both blendfake and deepfake effectively and comprehensively.

Cheng Yang, Rui Xu, Ye Guo et al.

Visual commonsense reasoning (VCR) is a challenging multi-modal task, which requires high-level cognition and commonsense reasoning ability about the real world. In recent years, large-scale pre-training approaches have been developed and promoted the state-of-the-art performance of VCR. However, the existing approaches almost employ the BERT-like objectives to learn multi-modal representations. These objectives motivated from the text-domain are insufficient for the excavation on the complex scenario of visual modality. Most importantly, the spatial distribution of the visual objects is basically neglected. To address the above issue, we propose to construct the spatial relation graph based on the given visual scenario. Further, we design two pre-training tasks named object position regression (OPR) and spatial relation classification (SRC) to learn to reconstruct the spatial relation graph respectively. Quantitative analysis suggests that the proposed method can guide the representations to maintain more spatial context and facilitate the attention on the essential visual regions for reasoning. We achieve the state-of-the-art results on VCR and two other vision-and-language reasoning tasks VQA, and NLVR.

Haojie Pan, Zepeng Zhai, Hao Yuan et al.

Driven by curiosity, humans have continually sought to explore and understand the world around them, leading to the invention of various tools to satiate this inquisitiveness. Despite not having the capacity to process and memorize vast amounts of information in their brains, humans excel in critical thinking, planning, reflection, and harnessing available tools to interact with and interpret the world, enabling them to find answers efficiently. The recent advancements in large language models (LLMs) suggest that machines might also possess the aforementioned human-like capabilities, allowing them to exhibit powerful abilities even with a constrained parameter count. In this paper, we introduce KwaiAgents, a generalized information-seeking agent system based on LLMs. Within KwaiAgents, we propose an agent system that employs LLMs as its cognitive core, which is capable of understanding a user's query, behavior guidelines, and referencing external documents. The agent can also update and retrieve information from its internal memory, plan and execute actions using a time-aware search-browse toolkit, and ultimately provide a comprehensive response. We further investigate the system's performance when powered by LLMs less advanced than GPT-4, and introduce the Meta-Agent Tuning (MAT) framework, designed to ensure even an open-sourced 7B or 13B model performs well among many agent systems. We exploit both benchmark and human evaluations to systematically validate these capabilities. Extensive experiments show the superiority of our agent system compared to other autonomous agents and highlight the enhanced generalized agent-abilities of our fine-tuned LLMs.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

Zhongyuan Wang, Richong Zhang, Zhijie Nie et al.

Recent Text-to-SQL methods leverage large language models (LLMs) by incorporating feedback from the database management system. While these methods effectively address execution errors in SQL queries, they struggle with database mismatches -- errors that do not trigger execution exceptions. Database mismatches include issues such as condition mismatches and stricter constraint mismatches, both of which are more prevalent in real-world scenarios. To address these challenges, we propose a tool-assisted agent framework for SQL inspection and refinement, equipping the LLM-based agent with two specialized tools: a retriever and a detector, designed to diagnose and correct SQL queries with database mismatches. These tools enhance the capability of LLMs to handle real-world queries more effectively. We also introduce Spider-Mismatch, a new dataset specifically constructed to reflect the condition mismatch problems encountered in real-world scenarios. Experimental results demonstrate that our method achieves the highest performance on the averaged results of the Spider and Spider-Realistic datasets in few-shot settings, and it significantly outperforms baseline methods on the more realistic dataset, Spider-Mismatch.

Yulan Hu, Sheng Ouyang, Jingyu Liu et al.

Graph contrastive learning (GCL) has emerged as a representative graph self-supervised method, achieving significant success. The currently prevalent optimization objective for GCL is InfoNCE. Typically, it employs augmentation techniques to obtain two views, where a node in one view acts as the anchor, the corresponding node in the other view serves as the positive sample, and all other nodes are regarded as negative samples. The goal is to minimize the distance between the anchor node and positive samples and maximize the distance to negative samples. However, due to the lack of label information during training, InfoNCE inevitably treats samples from the same class as negative samples, leading to the issue of false negative samples. This can impair the learned node representations and subsequently hinder performance in downstream tasks. While numerous methods have been proposed to mitigate the impact of false negatives, they still face various challenges. For instance, while increasing the number of negative samples can dilute the impact of false negatives, it concurrently increases computational burden. Thus, we propose GraphRank, a simple yet efficient graph contrastive learning method that addresses the problem of false negative samples by redefining the concept of negative samples to a certain extent, thereby avoiding the issue of false negative samples. The effectiveness of GraphRank is empirically validated through experiments on the node, edge, and graph level tasks.

Jikang Cheng, Jiaxin Ai, Zhen Han et al.

The face swapping technique based on deepfake methods poses significant social risks to personal identity security. While numerous deepfake detection methods have been proposed as countermeasures against malicious face swapping, they can only output binary labels (Fake/Real) for distinguishing fake content without reliable and traceable evidence. To achieve visual forensics and target face attribution, we propose a novel task named face retracing, which considers retracing the original target face from the given fake one via inverse mapping. Toward this goal, we propose an IDRetracor that can retrace arbitrary original target identities from fake faces generated by multiple face swapping methods. Specifically, we first adopt a mapping resolver to perceive the possible solution space of the original target face for the inverse mappings. Then, we propose mapping-aware convolutions to retrace the original target face from the fake one. Such convolutions contain multiple kernels that can be combined under the control of the mapping resolver to tackle different face swapping mappings dynamically. Extensive experiments demonstrate that the IDRetracor exhibits promising retracing performance from both quantitative and qualitative perspectives.

Ming Chen, Yan Zhou, Weihua Jian et al.

Though significant progress in human pose and shape recovery from monocular RGB images has been made in recent years, obtaining 3D human motion with high accuracy and temporal consistency from videos remains challenging. Existing video-based methods tend to reconstruct human motion from global image features, which lack detailed representation capability and limit the reconstruction accuracy. In this paper, we propose a Temporal-Aware Refining Network (TAR), to synchronously explore temporal-aware global and local image features for accurate pose and shape recovery. First, a global transformer encoder is introduced to obtain temporal global features from static feature sequences. Second, a bidirectional ConvGRU network takes the sequence of high-resolution feature maps as input, and outputs temporal local feature maps that maintain high resolution and capture the local motion of the human body. Finally, a recurrent refinement module iteratively updates estimated SMPL parameters by leveraging both global and local temporal information to achieve accurate and smooth results. Extensive experiments demonstrate that our TAR obtains more accurate results than previous state-of-the-art methods on popular benchmarks, i.e., 3DPW, MPI-INF-3DHP, and Human3.6M.

Zhangchi Feng, Dongdong Kuang, Zhongyuan Wang et al.

This paper presents EasyRAG, a simple, lightweight, and efficient retrieval-augmented generation framework for automated network operations. Our framework has three advantages. The first is accurate question answering. We designed a straightforward RAG scheme based on (1) a specific data processing workflow (2) dual-route sparse retrieval for coarse ranking (3) LLM Reranker for reranking (4) LLM answer generation and optimization. This approach achieved first place in the GLM4 track in the preliminary round and second place in the GLM4 track in the semifinals. The second is simple deployment. Our method primarily consists of BM25 retrieval and BGE-reranker reranking, requiring no fine-tuning of any models, occupying minimal VRAM, easy to deploy, and highly scalable; we provide a flexible code library with various search and generation strategies, facilitating custom process implementation. The last one is efficient inference. We designed an efficient inference acceleration scheme for the entire coarse ranking, reranking, and generation process that significantly reduces the inference latency of RAG while maintaining a good level of accuracy; each acceleration scheme can be plug-and-play into any component of the RAG process, consistently enhancing the efficiency of the RAG system. Our code and data are released at \url{https://github.com/BUAADreamer/EasyRAG}.

Jikang Cheng, Ying Zhang, Zhongyuan Wang et al.

Recent years have seen an increasing interest in physical adversarial attacks, which aim to craft deployable patterns for deceiving deep neural networks, especially for person detectors. However, the adversarial patterns of existing patch-based attacks heavily suffer from the self-coupling issue, where a degradation, caused by physical transformations, in any small patch segment can result in a complete adversarial dysfunction, leading to poor robustness in the complex real world. Upon this observation, we introduce the Decoupled adversarial Patch (DePatch) attack to address the self-coupling issue of adversarial patches. Specifically, we divide the adversarial patch into block-wise segments, and reduce the inter-dependency among these segments through randomly erasing out some segments during the optimization. We further introduce a border shifting operation and a progressive decoupling strategy to improve the overall attack capabilities. Extensive experiments demonstrate the superior performance of our method over other physical adversarial attacks, especially in the real world.

Huajie Tan, Yuheng Ji, Xiaoshuai Hao et al.

Visual reasoning abilities play a crucial role in understanding complex multimodal data, advancing both domain-specific applications and artificial general intelligence (AGI). Existing methods enhance Vision-Language Models (VLMs) through Chain-of-Thought (CoT) supervised fine-tuning using meticulously annotated data. However, this approach may lead to overfitting and cognitive rigidity, limiting the model's generalization ability under domain shifts and reducing real-world applicability. To overcome these limitations, we propose Reason-RFT, a two-stage reinforcement fine-tuning framework for visual reasoning. First, Supervised Fine-Tuning (SFT) with curated CoT data activates the reasoning potential of VLMs. This is followed by reinforcement learning based on Group Relative Policy Optimization (GRPO), which generates multiple reasoning-response pairs to enhance adaptability to domain shifts. To evaluate Reason-RFT, we reconstructed a comprehensive dataset covering visual counting, structural perception, and spatial transformation, serving as a benchmark for systematic assessment across three key dimensions. Experimental results highlight three advantages: (1) performance enhancement, with Reason-RFT achieving state-of-the-art results and outperforming both open-source and proprietary models; (2) generalization superiority, maintaining robust performance under domain shifts across various tasks; and (3) data efficiency, excelling in few-shot learning scenarios and surpassing full-dataset SFT baselines. Reason-RFT introduces a novel training paradigm for visual reasoning and marks a significant step forward in multimodal research. Project website: https://tanhuajie.github.io/ReasonRFT

Zezheng Wang, Zitong Yu, Xun Wang et al.

Face anti-spoofing (FAS) plays a crucial role in securing face recognition systems. Empirically, given an image, a model with more consistent output on different views of this image usually performs better, as shown in Fig.1. Motivated by this exciting observation, we conjecture that encouraging feature consistency of different views may be a promising way to boost FAS models. In this paper, we explore this way thoroughly by enhancing both Embedding-level and Prediction-level Consistency Regularization (EPCR) in FAS. Specifically, at the embedding-level, we design a dense similarity loss to maximize the similarities between all positions of two intermediate feature maps in a self-supervised fashion; while at the prediction-level, we optimize the mean square error between the predictions of two views. Notably, our EPCR is free of annotations and can directly integrate into semi-supervised learning schemes. Considering different application scenarios, we further design five diverse semi-supervised protocols to measure semi-supervised FAS techniques. We conduct extensive experiments to show that EPCR can significantly improve the performance of several supervised and semi-supervised tasks on benchmark datasets. The codes and protocols will be released at https://github.com/clks-wzz/EPCR.

Xing Cheng, Hezheng Lin, Xiangyu Wu et al.

The task of multi-label image classification is to recognize all the object labels presented in an image. Though advancing for years, small objects, similar objects and objects with high conditional probability are still the main bottlenecks of previous convolutional neural network(CNN) based models, limited by convolutional kernels' representational capacity. Recent vision transformer networks utilize the self-attention mechanism to extract the feature of pixel granularity, which expresses richer local semantic information, while is insufficient for mining global spatial dependence. In this paper, we point out the three crucial problems that CNN-based methods encounter and explore the possibility of conducting specific transformer modules to settle them. We put forward a Multi-label Transformer architecture(MlTr) constructed with windows partitioning, in-window pixel attention, cross-window attention, particularly improving the performance of multi-label image classification tasks. The proposed MlTr shows state-of-the-art results on various prevalent multi-label datasets such as MS-COCO, Pascal-VOC, and NUS-WIDE with 88.5%, 95.8%, and 65.5% respectively. The code will be available soon at https://github.com/starmemda/MlTr/

Hezheng Lin, Xing Cheng, Xiangyu Wu et al.

Since Transformer has found widespread use in NLP, the potential of Transformer in CV has been realized and has inspired many new approaches. However, the computation required for replacing word tokens with image patches for Transformer after the tokenization of the image is vast(e.g., ViT), which bottlenecks model training and inference. In this paper, we propose a new attention mechanism in Transformer termed Cross Attention, which alternates attention inner the image patch instead of the whole image to capture local information and apply attention between image patches which are divided from single-channel feature maps capture global information. Both operations have less computation than standard self-attention in Transformer. By alternately applying attention inner patch and between patches, we implement cross attention to maintain the performance with lower computational cost and build a hierarchical network called Cross Attention Transformer(CAT) for other vision tasks. Our base model achieves state-of-the-arts on ImageNet-1K, and improves the performance of other methods on COCO and ADE20K, illustrating that our network has the potential to serve as general backbones. The code and models are available at \url{https://github.com/linhezheng19/CAT}.

Kui Jiang, Zhongyuan Wang, Zheng Wang et al.

Low-light image enhancement aims to improve an image's visibility while keeping its visual naturalness. Different from existing methods tending to accomplish the relighting task directly by ignoring the fidelity and naturalness recovery, we investigate the intrinsic degradation and relight the low-light image while refining the details and color in two steps. Inspired by the color image formulation (diffuse illumination color plus environment illumination color), we first estimate the degradation from low-light inputs to simulate the distortion of environment illumination color, and then refine the content to recover the loss of diffuse illumination color. To this end, we propose a novel Degradation-to-Refinement Generation Network (DRGN). Its distinctive features can be summarized as 1) A novel two-step generation network for degradation learning and content refinement. It is not only superior to one-step methods, but also capable of synthesizing sufficient paired samples to benefit the model training; 2) A multi-resolution fusion network to represent the target information (degradation or contents) in a multi-scale cooperative manner, which is more effective to address the complex unmixing problems. Extensive experiments on both the enhancement task and joint detection task have verified the effectiveness and efficiency of our proposed method, surpassing the SOTA by \textit{0.70dB on average and 3.18\% in mAP}, respectively. The code is available at \url{https://github.com/kuijiang0802/DRGN}.

Baojin Huang, Zhongyuan Wang, Guangcheng Wang et al.

The existing face recognition datasets usually lack occlusion samples, which hinders the development of face recognition. Especially during the COVID-19 coronavirus epidemic, wearing a mask has become an effective means of preventing the virus spread. Traditional CNN-based face recognition models trained on existing datasets are almost ineffective for heavy occlusion. To this end, we pioneer a simulated occlusion face recognition dataset. In particular, we first collect a variety of glasses and masks as occlusion, and randomly combine the occlusion attributes (occlusion objects, textures,and colors) to achieve a large number of more realistic occlusion types. We then cover them in the proper position of the face image with the normal occlusion habit. Furthermore, we reasonably combine original normal face images and occluded face images to form our final dataset, termed as Webface-OCC. It covers 804,704 face images of 10,575 subjects, with diverse occlusion types to ensure its diversity and stability. Extensive experiments on public datasets show that the ArcFace retrained by our dataset significantly outperforms the state-of-the-arts. Webface-OCC is available at https://github.com/Baojin-Huang/Webface-OCC.

Tao Lu, Yuanzhi Wang, Yanduo Zhang et al.

Recently, convolutional neural networks (CNNs) have been widely employed to promote the face hallucination due to the ability to predict high-frequency details from a large number of samples. However, most of them fail to take into account the overall facial profile and fine texture details simultaneously, resulting in reduced naturalness and fidelity of the reconstructed face, and further impairing the performance of downstream tasks (e.g., face detection, facial recognition). To tackle this issue, we propose a novel external-internal split attention group (ESAG), which encompasses two paths responsible for facial structure information and facial texture details, respectively. By fusing the features from these two paths, the consistency of facial structure and the fidelity of facial details are strengthened at the same time. Then, we propose a split-attention in split-attention network (SISN) to reconstruct photorealistic high-resolution facial images by cascading several ESAGs. Experimental results on face hallucination and face recognition unveil that the proposed method not only significantly improves the clarity of hallucinated faces, but also encourages the subsequent face recognition performance substantially. Codes have been released at https://github.com/mdswyz/SISN-Face-Hallucination.

Kui Jiang, Zhongyuan Wang, Peng Yi et al.

Rain streaks in the air appear in various blurring degrees and resolutions due to different distances from their positions to the camera. Similar rain patterns are visible in a rain image as well as its multi-scale (or multi-resolution) versions, which makes it possible to exploit such complementary information for rain streak representation. In this work, we explore the multi-scale collaborative representation for rain streaks from the perspective of input image scales and hierarchical deep features in a unified framework, termed multi-scale progressive fusion network (MSPFN) for single image rain streak removal. For similar rain streaks at different positions, we employ recurrent calculation to capture the global texture, thus allowing to explore the complementary and redundant information at the spatial dimension to characterize target rain streaks. Besides, we construct multi-scale pyramid structure, and further introduce the attention mechanism to guide the fine fusion of this correlated information from different scales. This multi-scale progressive fusion strategy not only promotes the cooperative representation, but also boosts the end-to-end training. Our proposed method is extensively evaluated on several benchmark datasets and achieves state-of-the-art results. Moreover, we conduct experiments on joint deraining, detection, and segmentation tasks, and inspire a new research direction of vision task-driven image deraining. The source code is available at \url{https://github.com/kuihua/MSPFN}.

Zhongyuan Wang, Guangcheng Wang, Baojin Huang et al.

In order to effectively prevent the spread of COVID-19 virus, almost everyone wears a mask during coronavirus epidemic. This almost makes conventional facial recognition technology ineffective in many cases, such as community access control, face access control, facial attendance, facial security checks at train stations, etc. Therefore, it is very urgent to improve the recognition performance of the existing face recognition technology on the masked faces. Most current advanced face recognition approaches are designed based on deep learning, which depend on a large number of face samples. However, at present, there are no publicly available masked face recognition datasets. To this end, this work proposes three types of masked face datasets, including Masked Face Detection Dataset (MFDD), Real-world Masked Face Recognition Dataset (RMFRD) and Simulated Masked Face Recognition Dataset (SMFRD). Among them, to the best of our knowledge, RMFRD is currently theworld's largest real-world masked face dataset. These datasets are freely available to industry and academia, based on which various applications on masked faces can be developed. The multi-granularity masked face recognition model we developed achieves 95% accuracy, exceeding the results reported by the industry. Our datasets are available at: https://github.com/X-zhangyang/Real-World-Masked-Face-Dataset.

Sixian Zhang, Bohan Wang, Junqiang Wu et al.

Current metrics for text-to-image models typically rely on statistical metrics which inadequately represent the real preference of humans. Although recent work attempts to learn these preferences via human annotated images, they reduce the rich tapestry of human preference to a single overall score. However, the preference results vary when humans evaluate images with different aspects. Therefore, to learn the multi-dimensional human preferences, we propose the Multi-dimensional Preference Score (MPS), the first multi-dimensional preference scoring model for the evaluation of text-to-image models. The MPS introduces the preference condition module upon CLIP model to learn these diverse preferences. It is trained based on our Multi-dimensional Human Preference (MHP) Dataset, which comprises 918,315 human preference choices across four dimensions (i.e., aesthetics, semantic alignment, detail quality and overall assessment) on 607,541 images. The images are generated by a wide range of latest text-to-image models. The MPS outperforms existing scoring methods across 3 datasets in 4 dimensions, enabling it a promising metric for evaluating and improving text-to-image generation.

Siqi Fan, Xin Jiang, Xiang Li et al.

Due to the large number of parameters, the inference phase of Large Language Models (LLMs) is resource-intensive. However, not all requests posed to LLMs are equally difficult to handle. Through analysis, we show that for some tasks, LLMs can achieve results comparable to the final output at some intermediate layers. That is, not all layers of LLMs are necessary during inference. If we can predict at which layer the inferred results match the final results (produced by evaluating all layers), we could significantly reduce the inference cost. To this end, we propose a simple yet effective algorithm named AdaInfer to adaptively terminate the inference process for an input instance. AdaInfer relies on easily obtainable statistical features and classic classifiers like SVM. Experiments on well-known LLMs like the Llama2 series and OPT, show that AdaInfer can achieve an average of 17.8% pruning ratio, and up to 43% on sentiment tasks, with nearly no performance drop (<1%). Because AdaInfer does not alter LLM parameters, the LLMs incorporated with AdaInfer maintain generalizability across tasks.

Jiazhao Zhang, Kunyu Wang, Shaoan Wang et al.

A practical navigation agent must be capable of handling a wide range of interaction demands, such as following instructions, searching objects, answering questions, tracking people, and more. Existing models for embodied navigation fall short of serving as practical generalists in the real world, as they are often constrained by specific task configurations or pre-defined maps with discretized waypoints. In this work, we present Uni-NaVid, the first video-based vision-language-action (VLA) model designed to unify diverse embodied navigation tasks and enable seamless navigation for mixed long-horizon tasks in unseen real-world environments. Uni-NaVid achieves this by harmonizing the input and output data configurations for all commonly used embodied navigation tasks and thereby integrating all tasks in one model. For training Uni-NaVid, we collect 3.6 million navigation data samples in total from four essential navigation sub-tasks and foster synergy in learning across them. Extensive experiments on comprehensive navigation benchmarks clearly demonstrate the advantages of unification modeling in Uni-NaVid and show it achieves state-of-the-art performance. Additionally, real-world experiments confirm the model's effectiveness and efficiency, shedding light on its strong generalizability.

Qifan Liang, Junlin Li, Zhen Han et al.

Electrocautery or lasers will inevitably generate surgical smoke, which hinders the visual guidance of laparoscopic videos for surgical procedures. The surgical smoke can be classified into different types based on its motion patterns, leading to distinctive spatio-temporal characteristics across smoky laparoscopic videos. However, existing desmoking methods fail to account for such smoke-type-specific distinctions. Therefore, we propose the first Smoke-Type-Aware Laparoscopic Video Desmoking Network (STANet) by introducing two smoke types: Diffusion Smoke and Ambient Smoke. Specifically, a smoke mask segmentation sub-network is designed to jointly conduct smoke mask and smoke type predictions based on the attention-weighted mask aggregation, while a smokeless video reconstruction sub-network is proposed to perform specially desmoking on smoky features guided by two types of smoke mask. To address the entanglement challenges of two smoke types, we further embed a coarse-to-fine disentanglement module into the mask segmentation sub-network, which yields more accurate disentangled masks through the smoke-type-aware cross attention between non-entangled and entangled regions. In addition, we also construct the first large-scale synthetic video desmoking dataset with smoke type annotations. Extensive experiments demonstrate that our method not only outperforms state-of-the-art approaches in quality evaluations, but also exhibits superior generalization across multiple downstream surgical tasks.