Resolution generalization in image generation tasks enables the production of higher-resolution images with lower training resolution overhead. However, a key obstacle for diffusion transformers in addressing this problem is the mismatch between positional encodings seen at inference and those used during training. Existing strategies such as positional encodings interpolation, extrapolation, or hybrids, do not fully resolve this mismatch. In this paper, we propose a novel two-dimensional randomized positional encodings, namely RPE-2D, that prioritizes the order of image patches rather than their absolute distances, enabling seamless high- and low-resolution generation without training on multiple resolutions. Concretely, RPE-2D independently samples positions along the horizontal and vertical axes over an expanded range during training, ensuring that the encodings used at inference lie within the training distribution and thereby improving resolution generalization. We further introduce a simple random resize-and-crop augmentation to strengthen order modeling and add micro-conditioning to indicate the applied cropping pattern. On the ImageNet dataset, RPE-2D achieves state-of-the-art resolution generalization performance, outperforming competitive methods when trained at 256^2 and evaluated at 384^2 and 512^2, and when trained at 512^2 and evaluated at 768^2 and 1024^2. RPE-2D also exhibits outstanding capabilities in low-resolution image generation, multi-stage training acceleration, and multi-resolution inheritance.

Current video generation models perform well at single-shot synthesis but struggle with multi-shot videos, facing critical challenges in maintaining character and background consistency across shots and flexibly generating videos of arbitrary length and shot count. To address these limitations, we introduce \textbf{FilmWeaver}, a novel framework designed to generate consistent, multi-shot videos of arbitrary length. First, it employs an autoregressive diffusion paradigm to achieve arbitrary-length video generation. To address the challenge of consistency, our key insight is to decouple the problem into inter-shot consistency and intra-shot coherence. We achieve this through a dual-level cache mechanism: a shot memory caches keyframes from preceding shots to maintain character and scene identity, while a temporal memory retains a history of frames from the current shot to ensure smooth, continuous motion. The proposed framework allows for flexible, multi-round user interaction to create multi-shot videos. Furthermore, due to this decoupled design, our method demonstrates high versatility by supporting downstream tasks such as multi-concept injection and video extension. To facilitate the training of our consistency-aware method, we also developed a comprehensive pipeline to construct a high-quality multi-shot video dataset. Extensive experimental results demonstrate that our method surpasses existing approaches on metrics for both consistency and aesthetic quality, opening up new possibilities for creating more consistent, controllable, and narrative-driven video content.

Video large language models have achieved remarkable performance in tasks such as video question answering, however, their temporal understanding remains suboptimal. To address this limitation, we curate a dedicated instruction fine-tuning dataset that focuses on enhancing temporal comprehension across five key dimensions. In order to reduce reliance on costly temporal annotations, we introduce a multi-task prompt fine-tuning approach that seamlessly integrates temporal-sensitive tasks into existing instruction datasets without requiring additional annotations. Furthermore, we develop a novel benchmark for temporal-sensitive video understanding that not only fills the gaps in dimension coverage left by existing benchmarks but also rigorously filters out potential shortcuts, ensuring a more accurate evaluation. Extensive experimental results demonstrate that our approach significantly enhances the temporal understanding of video-LLMs while avoiding reliance on shortcuts.

Molecule discovery is a pivotal research field, impacting everything from medicine to materials. Recently, Large Language Models (LLMs) have been widely adopted in molecular understanding and generation, serving as a bridge between the molecular space and the natural language space, yet the alignment between molecules and their corresponding captions remains a significant challenge. Previous endeavors typically treat molecules as monolithic inputs, lacking an intermediate reasoning process and sacrificing explainability. In this work, we define fine-grained alignments as the precise correspondence between a molecule's sub-structures and the textual phrases that explain their properties. These alignments are crucial for LLMs to understand molecules in a more accurate and explainable manner. Normally, such fine-grained alignments require expert annotation, which is both costly and time-consuming. To allow LLMs to automatically label and learn the fine-grained alignments, we propose MolReFlect, a novel teacher-student framework, where a teacher LLM first generates and refines mappings between caption phrases and SMILES substructures and then explicitly teaches these detailed alignments to a student LLM. Experimental results demonstrate that MolReFlect enables LLMs to significantly outperform previous baselines, achieving the state-of-the-art performance in the molecule-caption translation task. Our codes are available via: https://github.com/phenixace/MolReFlect.

Large Language Models (LLMs) have achieved remarkable success and have been applied across various scientific fields, including chemistry. However, many chemical tasks require the processing of visual information, which cannot be successfully handled by existing chemical LLMs. This brings a growing need for models capable of integrating multimodal information in the chemical domain. In this paper, we introduce ChemVLM, an open-source chemical multimodal large language model specifically designed for chemical applications. ChemVLM is trained on a carefully curated bilingual multimodal dataset that enhances its ability to understand both textual and visual chemical information, including molecular structures, reactions, and chemistry examination questions. We develop three datasets for comprehensive evaluation, tailored to Chemical Optical Character Recognition (OCR), Multimodal Chemical Reasoning (MMCR), and Multimodal Molecule Understanding tasks. We benchmark ChemVLM against a range of open-source and proprietary multimodal large language models on various tasks. Experimental results demonstrate that ChemVLM achieves competitive performance across all evaluated tasks.

Recent Multi-modal Large Language Models (MLLMs) have made great progress in video understanding. However, their performance on videos involving human actions is still limited by the lack of high-quality data. To address this, we introduce a two-stage data annotation pipeline. First, we design strategies to accumulate videos featuring clear human actions from the Internet. Second, videos are annotated in a standardized caption format that uses human attributes to distinguish individuals and chronologically details their actions and interactions. Through this pipeline, we curate two datasets, namely HAICTrain and HAICBench. HAICTrain comprises 126K video-caption pairs generated by Gemini-Pro and verified for training purposes. Meanwhile, HAICBench includes 412 manually annotated video-caption pairs and 2,000 QA pairs, for a comprehensive evaluation of human action understanding. Experimental results demonstrate that training with HAICTrain not only significantly enhances human understanding abilities across 4 benchmarks, but can also improve text-to-video generation results. Both the HAICTrain and HAICBench will be made open-source to facilitate further research.

Recently, Large language models (LLMs) have revolutionized Natural Language Processing (NLP). Pretrained LLMs, due to limited training context size, struggle with handling long token sequences, limiting their performance on various downstream tasks. Current solutions toward long context modeling often employ multi-stage continual pertaining, which progressively increases the effective context length through several continual pretraining stages. However, those approaches require extensive manual tuning and human expertise. In this paper, we introduce a novel single-stage continual pretraining method, Head-Adaptive Rotary Position Embedding (HARPE), to equip LLMs with long context modeling capabilities while simplifying the training process. Our HARPE leverages different Rotary Position Embedding (RoPE) base frequency values across different attention heads and directly trains LLMs on the target context length. Extensive experiments on 4 language modeling benchmarks, including the latest RULER benchmark, demonstrate that HARPE excels in understanding and integrating long-context tasks with single-stage training, matching and even outperforming existing multi-stage methods. Our results highlight that HARPE successfully breaks the stage barrier for training LLMs with long context modeling capabilities.

Pre-trained video generation models hold great potential for generative video super-resolution (VSR). However, adapting them for full-size VSR, as most existing methods do, suffers from unnecessary intensive full-attention computation and fixed output resolution. To overcome these limitations, we make the first exploration into utilizing video diffusion priors for patch-wise VSR. This is non-trivial because pre-trained video diffusion models are not native for patch-level detail generation. To mitigate this challenge, we propose an innovative approach, called PatchVSR, which integrates a dual-stream adapter for conditional guidance. The patch branch extracts features from input patches to maintain content fidelity while the global branch extracts context features from the resized full video to bridge the generation gap caused by incomplete semantics of patches. Particularly, we also inject the patch’s location information into the model to better contextualize patch synthesis within the global video frame. Experiments demonstrate that our method can synthesize high-fidelity, high-resolution details at the patch level. A tailor-made multi-patch joint modulation is proposed to ensure visual consistency across individually enhanced patches. Due to the flexibility of our patch-based paradigm, we can achieve highly competitive 4K VSR based on a 512×512 resolution base model, with extremely high efficiency.

Existing radiance field-based head avatar methods have mostly relied on pre-computed explicit priors (e.g., mesh, point) or neural implicit representations, making it challenging to achieve high fidelity with both computational efficiency and low memory consumption. To overcome this, we present GPAvatar, a novel and efficient Gaussian splatting-based method for reconstructing high-fidelity dynamic 3D head avatars from monocular videos. We extend Gaussians in 3D space to a high-dimensional embedding space encompassing Gaussian’s spatial position and avatar expression, enabling the representation of the head avatar with arbitrary pose and expression. To enable splatting-based rasterization, a linear transformation is learned to project each high-dimensional Gaussian back to the 3D space, which is sufficient to capture expression variations instead of using complex neural networks. Furthermore, we propose an adaptive densification strategy that dynamically allocates Gaussians to regions with high expression variance, improving the facial detail representation. Experimental results on three datasets show that our method outperforms existing state-of-the-art methods in rendering quality and speed while reducing memory usage in training and rendering.

Video generation and editing conditioned on text prompts or images have undergone significant advancements. However, challenges remain in accurately controlling global layout and geometry details solely by texts, and supporting motion control and local modification through images. In this paper, we aim to achieve sketch-based spatial and motion control for video generation and support fine-grained editing of real or synthetic videos. Based on the DiT video generation model, we propose a memory-efficient control structure with sketch control blocks that predict residual features of skipped DiT blocks. Sketches are drawn on one or two keyframes (at arbitrary time points) for easy interaction. To propagate such temporally sparse sketch conditions across all frames, we propose an inter-frame attention mechanism to analyze the relationship between the keyframes and each video frame. For sketch-based video editing, we design an additional video insertion module that maintains consistency between the newly edited content and the original video’s spatial feature and dynamic motion. During inference, we use latent fusion for the accurate preservation of unedited regions. Extensive experiments demonstrate that our SketchVideo achieves superior performance in controllable video generation and editing.

Realistic simulation of dynamic scenes requires accurately capturing diverse material properties and modeling complex object interactions grounded in physical principles. However, existing methods are constrained to basic material types with limited predictable parameters, making them insufficient to represent the complexity of real-world materials. We introduce PhysFlow, a novel approach that leverages multi-modal foundation models and video diffusion to achieve enhanced 4D dynamic scene simulation. Our method utilizes multi-modal models to identify material types and initialize material parameters through image queries, while simultaneously inferring 3D Gaussian splats for detailed scene representation. We further refine these material parameters using video diffusion with a differentiable Material Point Method (MPM) and optical flow guidance rather than render loss or Score Distillation Sampling (SDS) loss. This integrated framework enables accurate prediction and realistic simulation of dynamic interactions in real-world scenarios, advancing both accuracy and flexibility in physics-based simulations. Our code and data are available at https://zhuomanliu.github.io/PhysFlow

With the continuous progress of visual generation technologies, the scale of video datasets has grown exponentially. The quality of these datasets plays a pivotal role in the performance of video generation models. We assert that temporal splitting, detailed captions, and video quality filtering are three crucial determinants of dataset quality. However, existing datasets exhibit various limitations in these areas. To address these challenges, we introduce Koala-36M, a large-scale, high-quality video dataset featuring accurate temporal splitting, detailed captions, and superior video quality. The essence of our approach lies in improving the consistency between fine-grained conditions and video content. Specifically, we employ a linear classifier on probability distributions to enhance the accuracy of transition detection, ensuring better temporal consistency. We then provide structured captions for the splitted videos, with an average length of 200 words, to improve text-video alignment. Additionally, we develop a Video Training Suitability Score (VTSS) that integrates multiple sub-metrics, allowing us to filter high-quality videos from the original corpus. Finally, we incorporate several metrics into the training process of the generation model, further refining the fine-grained conditions. Our experiments demonstrate the effectiveness of our data processing pipeline and the quality of the proposed Koala-36M dataset. Our dataset and code have been released at https://koala36m.github.io/.

Large Vision-Language Models (LVLMs) have achieved significant progress in recent years. However, the expensive inference cost limits the realistic deployment of LVLMs. Some works find that visual tokens are redundant and compress tokens to reduce the inference cost. These works identify important non-redundant tokens as target tokens, then prune the remaining tokens (non-target tokens) or merge them into target tokens. However, target token identification faces the token importance-redundancy dilemma. Besides, token merging and pruning face a dilemma between disrupting target token information and losing non-target token information. To solve these problems, we propose a novel visual token compression scheme, named Libra-Merging. In target token identification, Libra-Merging selects the most important tokens from spatially discrete intervals, achieving a more robust token importance-redundancy trade-off than relying on a hyper-parameter. In token compression, when non-target tokens are dissimilar to target tokens, Libra-Merging does not merge them into the target tokens, thus avoiding disrupting target token information. Meanwhile, Libra-Merging condenses these non-target tokens into an information compensation token to prevent losing important non-target token information. Our method can serve as a plug-in for diverse LVLMs, and extensive experimental results demonstrate its effectiveness. The code will be publicly available at https://github.com/longrongyang/Libra-Merging.

Style control has been popular in video generation models. Existing methods often generate videos far from the given style, cause content leakage, and struggle to transfer one video to the desired style. Our first observation is that the style extraction stage matters, whereas existing methods emphasize global style but ignore local textures. In order to bring texture features while preventing content leakage, we filter content-related patches while retaining style ones based on prompt-patch similarity; for global style extraction, we generate a paired style dataset through model illusion to facilitate contrastive learning, which greatly enhances the absolute style consistency. Moreover, to fill in the image-to-video gap, we train a lightweight motion adapter on still videos, which implicitly enhances stylization extent, and enables our image-trained model to be seamlessly applied to videos. Benefited from these efforts, our approach, StyleMaster, not only achieves significant improvement in both style resemblance and temporal coherence, but also can easily generalize to video style transfer with a gray tile ControlNet. Extensive experiments and visualizations demonstrate that StyleMaster significantly outperforms competitors, effectively generating high-quality stylized videos that align with textual content and closely resemble the style of reference images.

Achieving optimal performance of video diffusion transformers within given data and compute budget is crucial due to their high training costs. This necessitates precisely determining the optimal model size and training hyperparameters before large-scale training. While scaling laws are employed in language models to predict performance, their existence and accurate derivation in visual generation models remain underexplored. In this paper, we systematically analyze scaling laws for video diffusion transformers and confirm their presence. Moreover, we discover that, unlike language models, video diffusion models are more sensitive to learning rate and batch size—two hyperparameters often not precisely modeled. To address this, we propose a new scaling law that predicts optimal hyperparameters for any model size and compute budget. Under these optimal settings, we achieve comparable performance and reduce inference costs by 40.1% compared to conventional scaling methods, within a compute budget of 1e10 TFlops. Furthermore, we establish a more generalized and precise relationship among validation loss, any model size, and compute budget. This enables performance prediction for non-optimal model sizes, which may also be appealed under practical inference cost constraints, achieving a better trade-off.

Vision-language models (VLMs) have shown remarkable advancements in multimodal reasoning tasks. However, they still often generate inaccurate or irrelevant responses due to issues like hallucinated image understandings or unrefined reasoning paths. To address these challenges, we introduce Critic-V, a novel framework inspired by the Actor-Critic paradigm to boost the reasoning capability of VLMs. This framework decouples the reasoning process and critic process by integrating two independent components: the Reasoner, which generates reasoning paths based on visual and textual inputs, and the Critic, which provides constructive critique to refine these paths. In this approach, the Reasoner generates reasoning responses according to text prompts, which can evolve iteratively as a policy based on feedback from the Critic. This interaction process was theoretically driven by a reinforcement learning framework where the Critic offers natural language critiques instead of scalar rewards, enabling more nuanced feedback to boost the Reasoner’s capability on complex reasoning tasks. The Critic model is trained using Direct Preference Optimization (DPO), leveraging a preference dataset of critiques ranked by Rule-based Reward (RBR) to enhance its critic capabilities. Evaluation results show that the Critic-V framework significantly outperforms existing methods, including GPT-4V, on 5 out of 8 benchmarks, especially regarding reasoning accuracy and efficiency. Combining a dynamic text-based policy for the Reasoner and constructive feedback from the preference-optimized Critic enables a more reliable and context-sensitive multimodal reasoning process. Our approach provides a promising solution to enhance the reliability of VLMs, improving their performance in real-world reasoning-heavy multimodal applications such as autonomous driving and embodied intelligence. Our data and code are released at https://github.com/kyrieLei/Critic-V.

Minxuan Lv, Zhenpeng Su, Leiyu Pan, Yizhe Xiong, Zijia Lin, Hui Chen, Wei Zhou, Jungong Han, Guiguang Ding, Wenwu Ou, Di Zhang, Kun Gai, Songlin Hu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Camera control has been actively studied in text or image conditioned video generation tasks. However, altering camera trajectories of a given video remains under-explored, despite its importance in the field of video creation. It is non-trivial due to the extra constraints of maintaining multiple-frame appearance and dynamic synchronization. To address this, we present ReCamMaster, a camera-controlled generative video re-rendering framework that reproduces the dynamic scene of an input video at novel camera trajectories. The core innovation lies in harnessing the generative capabilities of pre-trained text-to-video models through a simple yet powerful video conditioning mechanism--its capability is often overlooked in current research. To overcome the scarcity of qualified training data, we construct a comprehensive multi-camera synchronized video dataset using Unreal Engine 5, which is carefully curated to follow real-world filming characteristics, covering diverse scenes and camera movements. It helps the model generalize to in-the-wild videos. Lastly, we further improve the robustness to diverse inputs through a meticulously designed training strategy. Extensive experiments show that our method substantially outperforms existing state-of-the-art approaches. Our method also finds promising applications in video stabilization, super-resolution, and outpainting. Our code and dataset are publicly available at: https://github.com/KwaiVGI/ReCamMaster.

Creating high-quality, generalizable speech-driven 3D talking heads remains a persistent challenge. Previous methods achieve satisfactory results for fixed viewpoints and small-scale audio variations, but they struggle with large head rotations and out-of-distribution (OOD) audio. Moreover, they are constrained by the need for time-consuming, identity-specific training. We believe the core issue lies in the lack of sufficient 3D priors, which limits the extrapolation capabilities of synthesized talking heads. To address this, we propose GGTalker, which synthesizes talking heads through a combination of generalizable priors and identity-specific adaptation. We introduce a two-stage Prior-Adaptation training strategy to learn Gaussian head priors and adapt to individual characteristics. We train Audio-Expression and Expression-Visual priors to capture the universal patterns of lip movements and the general distribution of head textures. During the Customized Adaptation, individual speaking styles and texture details are precisely modeled. Additionally, we introduce a color MLP to generate fine-grained, motion-aligned textures and a Body Inpainter to blend rendered results with the background, producing indistinguishable, photorealistic video frames. Comprehensive experiments show that GGTalker achieves state-of-the-art performance in rendering quality, 3D consistency, lip-sync accuracy, and training efficiency.

Existing text-to-image diffusion models have demonstrated remarkable capabilities in generating high-quality images guided by textual prompts. However, achieving multi-subject compositional synthesis with precise spatial control remains a significant challenge. In this work, we address the task of layout-controllable multi-subject synthesis (LMS), which requires both faithful reconstruction of reference subjects and their accurate placement in specified regions within a unified image. While recent advancements have separately improved layout control and subject synthesis, existing approaches struggle to simultaneously satisfy the dual requirements of spatial precision and identity preservation in this composite task. To bridge this gap, we propose MUSE, a unified synthesis framework that employs concatenated cross-attention (CCA) to seamlessly integrate layout specifications with textual guidance through explicit semantic space expansion. The proposed CCA mechanism enables bidirectional modality alignment between spatial constraints and textual descriptions without interference. Furthermore, we design a progressive two-stage training strategy that decomposes the LMS task into learnable sub-objectives for effective optimization. Extensive experiments demonstrate that MUSE achieves zero-shot end-to-end generation with superior spatial accuracy and identity consistency compared to existing solutions, advancing the frontier of controllable image synthesis. Our code and model are available at https://github.com/pf0607/MUSE.

In visual generation tasks, the responses and combinations of complex concepts often lack stability and are error-prone, which remains an under-explored area. In this paper, we attempt to explore the causal factors for poor concept responses through elaborately designed experiments. We also design a concept-wise equalization loss function (IMBA loss) to address this issue. Our proposed method is online, eliminating the need for offline dataset processing, and requires minimal code changes. In our newly proposed complex concept benchmark Inert-CompBench and two other public test sets, our method significantly enhances the concept response capability of baseline models and yields highly competitive results with only a few codes released at https://github.com/KwaiVGI/IMBA-Loss.

Recent advancements in video diffusion models demonstrate remarkable capabilities in simulating real-world dynamics and 3D consistency. This progress motivates us to explore the potential of these models to maintain dynamic consistency across diverse viewpoints, a feature highly sought after in applications like virtual filming. Unlike existing methods focused on multi-view generation of single objects for 4D reconstruction, our interest lies in generating open-world videos from arbitrary viewpoints, incorporating six degrees of freedom (6 DoF) camera poses. To achieve this, we propose a plug-and-play module that enhances a pre-trained text-to-video model for multi-camera video generation, ensuring consistent content across different viewpoints. Specifically, we introduce a multi-view synchronization module designed to maintain appearance and geometry consistency across these viewpoints. Given the scarcity of high-quality training data, we also propose a progressive training scheme that leverages multi-camera images and monocular videos as a supplement to Unreal Engine-rendered multi-camera videos. This comprehensive approach significantly benefits our model. Experimental results demonstrate the superiority of our proposed method over existing competitors and several baselines. Furthermore, our method enables intriguing extensions, such as re-rendering a video from multiple novel viewpoints. Project webpage: https://jianhongbai.github.io/SynCamMaster/

Multimodal visual language models are gaining prominence in open-world applications, driven by advancements in model architectures, training techniques, and high-quality data. However, their performance is often limited by insufficient task-specific data, leading to poor generalization and biased outputs. Existing efforts to increase task diversity in fine-tuning datasets are hindered by the labor-intensive process of manual task labeling, which typically produces only a few hundred task types. To address this, we propose TaskGalaxy, a large-scale multimodal instruction fine-tuning dataset comprising 19,227 hierarchical task types and 413,648 samples. TaskGalaxy utilizes GPT-4o to enrich task diversity by expanding from a small set of manually defined tasks, with CLIP and GPT-4o filtering those that best match open-source images, and generating relevant question-answer pairs. Multiple models are employed to ensure sample quality. This automated process enhances both task diversity and data quality, reducing manual intervention. Incorporating TaskGalaxy into LLaVA-v1.5 and InternVL-Chat-v1.0 models shows substantial performance improvements across 16 benchmarks, demonstrating the critical importance of task diversity. TaskGalaxy is publicly released at https://github.com/Kwai-YuanQi/TaskGalaxy.

Speech-driven 3D talking face method should offer both accurate lip synchronization and controllable expressions. Previous methods solely adopt discrete emotion labels to globally control expressions throughout sequences while limiting flexible fine-grained facial control within the spatiotemporal domain. We propose a diffusion-transformer-based 3D talking face generation model, Cafe-Talk, which simultaneously incorporates coarse- and fine-grained multimodal control conditions. Nevertheless, the entanglement of multiple conditions challenges achieving satisfying performance. To disentangle speech audio and fine-grained conditions, we employ a two-stage training pipeline. Specifically, Cafe-Talk is initially trained using only speech audio and coarse-grained conditions. Then, a proposed fine-grained control adapter gradually adds fine-grained instructions represented by action units (AUs), preventing unfavorable speech-lip synchronization. To disentangle coarse- and fine-grained conditions, we design a swap-label training mechanism, which enables the dominance of the fine-grained conditions. We also devise a mask-based CFG technique to regulate the occurrence and intensity of fine-grained control. In addition, a text-based detector is introduced with text-AU alignment to enable natural language user input and further support multimodal control. Extensive experimental results prove that Cafe-Talk achieves state-of-the-art lip synchronization and expressiveness performance and receives wide acceptance in fine-grained control in user studies.

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. 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. Extensive experiments validate the effectiveness and advantages of our approach, underscoring Stable-SAM as a more robust solution for segmenting anything. Codes are at https://github.com/fanq15/Stable-SAM.

This paper aims to manipulate multi-entity 3D motions in video generation. Previous methods on controllable video generation primarily leverage 2D control signals to manipulate object motions and have achieved remarkable synthesis results. However, 2D control signals are inherently limited in expressing the 3D nature of object motions. To overcome this problem, we introduce 3DTrajMaster, a robust controller that regulates multi-entity dynamics in 3D space, given user-desired 6DoF pose (location and rotation) sequences of entities. At the core of our approach is a plug-and-play 3D-motion grounded object injector that fuses multiple input entities with their respective 3D trajectories through a gated self-attention mechanism. In addition, we exploit an injector architecture to preserve the video diffusion prior, which is crucial for generalization ability. To mitigate video quality degradation, we introduce a domain adaptor during training and employ an annealed sampling strategy during inference. To address the lack of suitable training data, we construct a 360-Motion Dataset, which first correlates collected 3D human and animal assets with GPT-generated trajectory and then captures their motion with 12 evenly-surround cameras on diverse 3D UE platforms. Extensive experiments show that 3DTrajMaster sets a new state-of-the-art in both accuracy and generalization for controlling multi-entity 3D motions. Project page: http://fuxiao0719.github.io/projects/3dtrajmaster

The Mixture-of-Experts (MoE) has gained increasing attention in studying Large Vision-Language Models (LVLMs). It uses a sparse model to replace the dense model, achieving comparable performance while activating fewer parameters during inference, thus significantly reducing the inference cost. Existing MoE methods in LVLM encourage different experts to specialize in different tokens, and they usually employ a router to predict the routing of each token. However, the router is not optimized concerning distinct parameter optimization directions generated from tokens within an expert. This may lead to severe interference between tokens within an expert. To address this problem, we propose to use the token-level gradient analysis to Solving Token Gradient Conflict (STGC) in this paper. Specifically, we first use token-level gradients to identify conflicting tokens in experts. After that, we add a regularization loss tailored to encourage conflicting tokens routing from their current experts to other experts, for reducing interference between tokens within an expert. Our method can serve as a plug-in for diverse LVLM methods, and extensive experimental results demonstrate its effectiveness. demonstrate its effectiveness. The code will be publicly available at https://github.com/longrongyang/STGC.

Existing efforts to align multimodal large language models (MLLMs) with human preferences have only achieved progress in narrow areas, such as hallucination reduction, but remain limited in practical applicability and generalizability. To this end, we introduce **MM-RLHF**, a dataset containing **120k** fine-grained, human-annotated preference comparison pairs. This dataset represents a substantial advancement over existing resources, offering superior size, diversity, annotation granularity, and quality. Leveraging this dataset, we propose several key innovations to improve both the quality of reward models and the efficiency of alignment algorithms. Notably, we introduce the **Critique-Based Reward Model**, which generates critiques of model outputs before assigning scores, offering enhanced interpretability and more informative feedback compared to traditional scalar reward mechanisms. Additionally, we propose **Dynamic Reward Scaling**, a method that adjusts the loss weight of each sample according to the reward signal, thereby optimizing the use of high-quality comparison pairs. Our approach is rigorously evaluated across **10** distinct dimensions, encompassing **27** benchmarks, with results demonstrating significant and consistent improvements in model performance (Figure.1).

Multimodal large language models (MLLMs) recently showed strong capacity in integrating data among multiple modalities, empowered by generalizable attention architecture. Advanced methods predominantly focus on language-centric tuning while less exploring multimodal tokens mixed through attention, posing challenges in high-level tasks that require fine-grained cognition and emotion understanding. In this work, we identify the attention deficit disorder problem in multimodal learning, caused by inconsistent cross-modal attention and layer-by-layer decayed attention activation. To address this, we propose a novel attention mechanism, termed MOdular Duplex Attention (MODA), simultaneously conducting the inner-modal refinement and inter-modal interaction. MODA employs a correct-after-align strategy to effectively decouple modality alignment from cross-layer token mixing. In the alignment phase, tokens are mapped to duplex modality spaces based on the basis vectors, enabling the interaction between visual and language modality. Further, the correctness of attention scores is ensured through adaptive masked attention, which enhances the model's flexibility by allowing customizable masking patterns for different modalities. Extensive experiments on 21 benchmark datasets verify the effectiveness of MODA in perception, cognition, and emotion tasks.

Traditional navigation systems often overlook dynamic social contexts, affecting users' sense of safety and comfort in unfamiliar settings. This challenge requires integrating personalized real-time social data into route planning on resource-constrained mobile and wearable platforms. This paper introduces AuraNav, enhancing navigation by incorporating real-time recognition and individualized social patterns. AuraNav includes two main modules: (i) Social Topology Enhanced Navigation uses real-time familiar individual detection to create dynamic safety corridors by adjusting path costs with an influence field-based metric in A* search; and (ii) Personalized Path Memory mines long-term familiarity data to build heatmaps, offering route recommendations aligned with user habits and comfort zones. This system uses a modular edge-cloud architecture for low-latency sensor data processing and scalable analysis of social-spatial information. The evaluation shows that AuraNav improves routing safety by 17-20% and reduces travel time by 5%, maintaining immediate responsiveness with negligible latency. AuraNav offers a framework for socially aware navigation systems that evolve into personalized and context-aware systems.

Long-context video understanding in Multimodal Large Language Models (MLLMs) faces a critical challenge: balancing computational efficiency with the retention of fine-grained spatio-temporal patterns. Existing approaches (e.g., sparse sampling, dense sampling with low resolution, and token compression) suffer from significant information loss in temporal dynamics, spatial details, or subtle interactions, particularly in videos with complex motion or varying resolutions. To address this, we propose Mavors, a novel framework that introduces Multi-granularity video representation for holistic long-video modeling. Specifically, Mavors directly encodes raw video content into latent representations through two core components: 1) an Intra-chunk Vision Encoder (IVE) that preserves high-resolution spatial features via 3D convolutions and Vision Transformers, and 2) an Inter-chunk Feature Aggregator (IFA) that establishes temporal coherence across chunks using transformer-based dependency modeling with chunk-level rotary position encodings. Moreover, the framework unifies image and video understanding by treating images as single-frame videos via sub-image decomposition. Experiments across diverse benchmarks demonstrate Mavors' superiority in maintaining both spatial fidelity and temporal continuity, significantly outperforming existing methods in tasks requiring fine-grained spatio-temporal reasoning.

Di Zhang, Jianbo Wu, Jingdi Lei, Tong Che, Jiatong Li, Tong Xie, Xiaoshui Huang, Shufei Zhang, Marco Pavone, Yuqiang Li, Wanli Ouyang, Dongzhan Zhou. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Although multimodal large language models (MLLMs) exhibit remarkable reasoning capabilities on complex multimodal understanding tasks, they still suffer from the notorious 'hallucination' issue: generating outputs misaligned with obvious visual or factual evidence. Currently, training-based solutions, like direct preference optimization (DPO), leverage paired preference data to suppress hallucinations. However, they risk sacrificing general reasoning capabilities due to the likelihood displacement. Meanwhile, training-free solutions, like contrastive decoding, achieve this goal by subtracting the estimated hallucination pattern from a distorted input. Yet, these handcrafted perturbations (e.g., add noise to images) may poorly capture authentic hallucination patterns. To avoid these weaknesses of existing methods, and realize ``robust'' hallucination mitigation (\ie, maintaining general reasoning performance), we propose a novel framework: Decoupling Contrastive Decoding (DCD). Specifically, DCD decouples the learning of positive and negative samples in preference datasets, and trains separate positive and negative image projections within the MLLM. The negative projection implicitly models real hallucination patterns, which enables vision-aware negative images in the contrastive decoding inference stage. Our DCD alleviates likelihood displacement by avoiding pairwise optimization and generalizes robustly without handcrafted degradation. Extensive ablations across hallucination benchmarks and general reasoning tasks demonstrate the effectiveness of DCD, \ie, it matches DPO’s hallucination suppression while preserving general capabilities and outperforms the handcrafted contrastive decoding methods.

We propose Flow-GRPO, the first method to integrate online policy gradient reinforcement learning (RL) into flow matching models. Our approach uses two key strategies: (1) an ODE-to-SDE conversion that transforms a deterministic Ordinary Differential Equation (ODE) into an equivalent Stochastic Differential Equation (SDE) that matches the original model's marginal distribution at all timesteps, enabling statistical sampling for RL exploration; and (2) a Denoising Reduction strategy that reduces training denoising steps while retaining the original number of inference steps, significantly improving sampling efficiency without sacrificing performance. Empirically, Flow-GRPO is effective across multiple text-to-image tasks. For compositional generation, RL-tuned SD3.5-M generates nearly perfect object counts, spatial relations, and fine-grained attributes, increasing GenEval accuracy from $63\%$ to $95\%$. In visual text rendering, accuracy improves from $59\%$ to $92\%$, greatly enhancing text generation. Flow-GRPO also achieves substantial gains in human preference alignment. Notably, very little reward hacking occurred, meaning rewards did not increase at the cost of appreciable image quality or diversity degradation.

Video generation has achieved significant advances through rectified flow techniques, but issues like unsmooth motion and misalignment between videos and prompts persist. In this work, we develop a systematic pipeline that harnesses human feedback to mitigate these problems and refine the video generation model. Specifically, we begin by constructing a large-scale human preference dataset focused on modern video generation models, incorporating pairwise annotations across multi-dimensions. We then introduce VideoReward, a multi-dimensional video reward model, and examine how annotations and various design choices impact its rewarding efficacy. From a unified reinforcement learning perspective aimed at maximizing reward with KL regularization, we introduce three alignment algorithms for flow-based models. These include two training-time strategies: direct preference optimization for flow (Flow-DPO) and reward weighted regression for flow (Flow-RWR), and an inference-time technique, Flow-NRG, which applies reward guidance directly to noisy videos. Experimental results indicate that VideoReward significantly outperforms existing reward models, and Flow-DPO demonstrates superior performance compared to both Flow-RWR and supervised fine-tuning methods. Additionally, Flow-NRG lets users assign custom weights to multiple objectives during inference, meeting personalized video quality needs.

Lip synchronization is the task of aligning a speaker’s lip movements in video with corresponding speech audio, and it is essential for creating realistic, expressive video content. However, existing methods often rely on reference frames and masked-frame inpainting, which limit their robustness to identity consistency, pose variations, facial occlusions, and stylized content. In addition, since audio signals provide weaker conditioning than visual cues, lip shape leakage from the original video will affect lip sync quality. In this paper, we present OmniSync, a universal lip synchronization framework for diverse visual scenarios. Our approach introduces a mask-free training paradigm using Diffusion Transformer models for direct frame editing without explicit masks, enabling unlimited-duration inference while maintaining natural facial dynamics and preserving character identity. During inference, we propose a flow-matching-based progressive noise initialization to ensure pose and identity consistency, while allowing precise mouth-region editing. To address the weak conditioning signal of audio, we develop a Dynamic Spatiotemporal Classifier-Free Guidance (DS-CFG) mechanism that adaptively adjusts guidance strength over time and space. We also establish the AIGC-LipSync Benchmark, the first evaluation suite for lip synchronization in diverse AI-generated videos. Extensive experiments demonstrate that OmniSync significantly outperforms prior methods in both visual quality and lip sync accuracy, achieving superior results in both real-world and AI-generated videos.

Retrieval-augmented generation (RAG) has become a widely adopted paradigm for enabling knowledge-grounded large language models (LLMs). However, standard RAG pipelines often fail to ensure that model reasoning remains consistent with the evidence retrieved, leading to factual inconsistencies or unsupported conclusions. In this work, we reinterpret RAG as \textit{Retrieval-Augmented Reasoning} and identify a central but underexplored problem: \textit{Reasoning Misalignment}—the divergence between an LLM's internal reasoning trajectory and the evidential constraints provided by retrieval. To address this issue, we propose \textsc{AlignRAG}, a novel iterative framework grounded in \textit{Critique-Driven Alignment (CDA)}. We further introduce \textsc{AlignRAG-auto}, an autonomous variant that dynamically terminates refinement, removing the need to pre-specify the number of critique iterations. At the heart of \textsc{AlignRAG} lies a \textit{contrastive critique synthesis} mechanism that generates retrieval-sensitive critiques while mitigating self-bias. This mechanism trains a dedicated retrieval-augmented \textit{Critic Language Model (CLM)} using labeled critiques that distinguish between evidence-aligned and misaligned reasoning. Empirical evaluations show that our approach significantly improves reasoning fidelity. Our 8B-parameter CLM improves performance over the Self-Refine baseline by \textbf{12.1\%} on out-of-domain tasks and outperforms a standard 72B-parameter CLM by \textbf{2.2\%}. Furthermore, \textsc{AlignRAG-auto} achieves this state-of-the-art performance while dynamically determining the optimal number of refinement steps, enhancing efficiency and usability. \textsc{AlignRAG} remains compatible with existing RAG architectures as a \textit{plug-and-play} module and demonstrates strong robustness under both informative and noisy retrieval scenarios. Overall, \textsc{AlignRAG} offers a principled solution for aligning model reasoning with retrieved evidence, substantially improving the factual reliability and robustness of RAG systems. Our source code is provided at \href{https://github.com/upup-wei/RAG-ReasonAlignment}{link}.

Preference optimization for diffusion models aims to align them with human preferences for images. Previous methods typically use Vision-Language Models (VLMs) as pixel-level reward models to approximate human preferences. However, when used for step-level preference optimization, these models face challenges in handling noisy images of different timesteps and require complex transformations into pixel space. In this work, we show that pre-trained diffusion models are naturally suited for step-level reward modeling in the noisy latent space, as they are explicitly designed to process latent images at various noise levels. Accordingly, we propose the **Latent Reward Model (LRM)**, which repurposes components of the diffusion model to predict preferences of latent images at arbitrary timesteps. Building on LRM, we introduce **Latent Preference Optimization (LPO)**, a step-level preference optimization method conducted directly in the noisy latent space. Experimental results indicate that LPO significantly improves the model's alignment with general, aesthetic, and text-image alignment preferences, while achieving a 2.5-28x training speedup over existing preference optimization methods.

Understanding and predicting emotions from videos has gathered significant attention in recent studies, driven by advancements in video large language models (VideoLLMs). While advanced methods have made progress in video emotion analysis, the intrinsic nature of emotions—characterized by their open-set, dynamic, and context-dependent properties—poses challenge in understanding complex and evolving emotional states with reasonable rationale. To tackle these challenges, we propose a novel affective cues-guided reasoning framework that unifies fundamental attribute perception, expression analysis, and high-level emotional understanding in a stage-wise manner. At the core of our approach is a family of video emotion foundation models (VidEmo), specifically designed for emotion reasoning and instruction-following. These models undergo a two-stage tuning process: first, curriculum emotion learning for injecting emotion knowledge, followed by affective-tree reinforcement learning for emotion reasoning. Moreover, we establish a foundational data infrastructure and introduce a emotion-centric fine-grained dataset (Emo-CFG) consisting of 2.1M diverse instruction-based samples. Emo-CFG includes explainable emotional question-answering, fine-grained captions, and associated rationales, providing essential resources for advancing emotion understanding tasks. Experimental results demonstrate that our approach achieves competitive performance, setting a new milestone across 15 face perception tasks.

Camera control is crucial for generating expressive and cinematic videos. Existing methods rely on explicit sequences of camera parameters as control conditions, which can be cumbersome for users to construct, particularly for intricate camera movements. To provide a more intuitive camera control method, we propose CamCloneMaster, a framework that enables users to replicate camera movements from reference videos without requiring camera parameters or test-time fine-tuning. CamCloneMaster seamlessly supports reference-based camera control for both Image-to-Video and Video-to-Video tasks within a unified framework. Furthermore, we present the Camera Clone Dataset, a large-scale synthetic dataset designed for camera clone learning, encompassing diverse scenes, subjects, and camera movements. Extensive experiments and user studies demonstrate that CamCloneMaster outperforms existing methods in terms of both camera controllability and visual quality. Dataset and Code can be found at https://camclonemaster.github.io/.

Recent advances in interactive video generation have shown promising results, yet existing approaches struggle with scene-consistent memory capabilities in long video generation due to limited use of historical context. In this work, we propose Context-as-Memory, which utilizes historical context as memory for video generation. It includes two simple yet effective designs: (1) storing context in frame format without additional post-processing; (2) conditioning by concatenating context and frames to be predicted along the frame dimension at the input, requiring no external control modules. Furthermore, considering the enormous computational overhead of incorporating all historical context, we propose the Memory Retrieval module to select truly relevant context frames by determining FOV (Field of View) overlap between camera poses, which significantly reduces the number of candidate frames without substantial information loss. Experiments demonstrate that Context-as-Memory achieves superior memory capabilities in interactive long video generation compared to SOTAs, even generalizing effectively to open-domain scenarios not seen during training. Our project page are publicly available at https://context-as-memory.github.io/.

Recently, video-language understanding has achieved great success through large-scale pre-training. However, data scarcity remains a prevailing challenge. This study quantitatively reveals an "impossible trinity" among data quantity, diversity, and quality in pre-training datasets. Recent efforts seek to refine large-scale, diverse ASR datasets compromised by low quality through synthetic annotations. These methods successfully refine the original annotations by leveraging useful information in multimodal video content (frames, tags, ASR transcripts, etc.). Nevertheless, they struggle to mitigate noise within synthetic annotations and lack scalability as the dataset size expands. To address these issues, we introduce the Video DataFlywheel framework, which iteratively refines video annotations with improved noise control methods. For iterative refinement, we first leverage a video-language model to generate synthetic annotations, resulting in a refined dataset. Then, we pre-train on it and fine-tune on human refinement examples for a stronger model. These processes are repeated for continuous improvement. For noise control, we present AdaTaiLr, a novel method that requires weaker assumptions on noise distribution. This method proves more effective in large datasets and offers theoretical guarantees. The combination of iterative refinement and AdaTaiLr can achieve better scalability in video-language understanding. Extensive experiments show that our framework outperforms existing data refinement baselines, delivering a 3% performance boost and improving dataset quality with minimal diversity loss. Furthermore, our refined dataset facilitates significant improvements in various video-language understanding tasks, including video question answering and text-video retrieval.

The stable diffusion model (SDM) is a prevalent and effective model for text-to-image (T2I) and image-to-image (I2I) generation. Despite various attempts at sampler optimization, model distillation, and network quantification, these approaches typically maintain the original network architecture. The extensive parameter scale and substantial computational demands have limited research into adjusting the model architecture. This study focuses on reducing redundant computation in SDM and optimizes the model through both tuning and tuning-free methods: 1) for the tuning method, we design a model assembly strategy to reconstruct a lightweight model while preserving performance and ensuring semantic stability through distillation and 2) for the tuning-free method, we propose a feature inheritance strategy to accelerate inference by skipping local computations at the block, layer, or unit level within the network structure. We also examine multiple sampling modes for feature inheritance at the time-step level. Experiments demonstrate that both the proposed tuning and the tuning-free methods can improve the speed and performance of the SDM. The lightweight model reconstructed by the model assembly strategy increases generation speed by 22.4%, while the feature inheritance strategy enhances the SDM generation speed by 40.0%.

Yuchong Sun, Che Liu, Kun Zhou, Jinwen Huang, Ruihua Song, Xin Zhao, Fuzheng Zhang, Di Zhang, Kun Gai. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Large-scale GPU clusters have been widely used for effectively training both online and offline deep learning (DL) jobs. However, elastic scheduling in most cases of resource schedulers is dedicated for offline model training where resource adjustment is planned ahead of time. The native autoscaling policy is on the basis of pre-defined threshold and, if applied directly in online model training, often suffers from belated resource adjustment, leading to diminished model accuracy. In this paper, we present Kale, a novel elastic GPU scheduling system to improve the performance of online DL model training. Through traffic forecasting and resource-throughput modeling, Kale automatically pinpoints the number of required GPUs that best accommodate the on-the-fly data samples before performing stabilized autoscaling. An advanced data shuffling strategy is further employed for balancing uneven samples among different training workers, thereby improving the runtime efficacy. Experiments show that Kale substantially outperforms the state-of-the-art solutions. Compared with the default HPA autoscaling strategy, Kale reduces the accumulated lag and downtime by 69.2% and 33.1%, respectively, whilst lowering the SLO violation rate from 19.57% to just 2.6%. Kale has been deployed at Kuaishou's production-level GPU clusters and successfully underpins real-time video recommendation and advertisement at scale.

Large language models have demonstrated exceptional capability in natural language understanding and generation. However, their generation speed is limited by the inherently sequential nature of their decoding process, posing challenges for real-time applications. This paper introduces Lexical Unit Decoding (LUD), a novel decoding methodology implemented in a data-driven manner, accelerating the decoding process without sacrificing output quality. The core of our approach is the observation that a pre-trained language model can confidently predict multiple contiguous tokens, forming the basis for a lexical unit, in which these contiguous tokens could be decoded in parallel. Extensive experiments validate that our method substantially reduces decoding time while maintaining generation quality, i.e., 33% speed up on natural language generation with no quality loss, and 30% speed up on code generation with a negligible quality loss of 3%. Distinctively, LUD requires no auxiliary models and does not require changes to existing architectures. It can also be integrated with other decoding acceleration methods, thus achieving an even more pronounced inference efficiency boost. We posit that the foundational principles of LUD could define a new decoding paradigm for future language models, enhancing their applicability for a broader spectrum of applications. All codes are be publicly available at https://github.com/tjunlp-lab/Lexical-Unit-Decoding-LUD-.

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 multidimensional human preferences, we propose the Multi-dimensional Preference Score (MPS), the first multidimensional 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. The model and dataset will be made publicly available to facilitate future research. Project page: htt ps: //wangbohan97.github.io/MPS/.

Hallucination detection is a challenging task for large language models (LLMs), and existing studies heavily rely on powerful closed-source LLMs such as GPT-4.In this paper, we propose an autonomous LLM-based agent framework, called HaluAgent, which enables relatively smaller LLMs 1 (e.g., Baichuan2-Chat 7B) to actively select suitable tools for detecting multiple hallucination types such as text, code, and mathematical expression.In HaluAgent, we integrate the LLM, multi-functional toolbox, and design a fine-grained three-stage detection framework along with memory mechanism.To facilitate the effectiveness of HaluAgent, we leverage existing Chinese and English datasets to synthesize detection trajectories for fine-tuning, which endows HaluAgent with the capability for bilingual hallucination detection.Extensive experiments demonstrate that only using 2K samples for tuning LLMs, HaluAgent can perform hallucination detection on various types of tasks and datasets, achieving performance comparable to or even higher than GPT-4 without tool enhancements on both indomain and out-of-domain datasets.We release our dataset and code at https://github.com/ RUCAIBox/HaluAgent.

With the growing popularity of general-purpose Large Language Models (LLMs), comes a need for more global explanations of model behaviors.Concept-based explanations arise as a promising avenue for explaining high-level patterns learned by LLMs.Yet their evaluation poses unique challenges, especially due to their non-local nature and high dimensional representation in a model's hidden space.Current methods approach concepts from different perspectives, lacking a unified formalization.This makes evaluating the core measures of concepts, namely faithfulness or readability, challenging.To bridge the gap, we introduce a formal definition of concepts generalizing to diverse concept-based explanations' settings.Based on this, we quantify the faithfulness of a concept explanation via perturbation.We ensure adequate perturbation in the highdimensional space for different concepts via an optimization problem.Readability is approximated via an automatic and deterministic measure, quantifying the coherence of patterns that maximally activate a concept while aligning with human understanding.Finally, based on measurement theory, we apply a metaevaluation method for evaluating these measures, generalizable to other types of explanations or tasks as well.Extensive experimental analysis has been conducted to inform the selection of explanation evaluation measures.

Aligning large language models (LLMs) with human expectations requires high-quality instructional dialogues, which usually require instructions that are diverse and in-depth.Existing methods leverage two LLMs to interact for automatic collection: one simulating a user to pose instructions, and the other acting as a system agent to respond.However, these user simulators struggle to model the rules behind how dialogues can pose different instructions without explicit guidance, resulting in general instructions.In this paper, we propose to explicitly capture the complex rules to help the user simulator pose diverse and in-depth instruction.Specifically, we first induce high-level instruction strategies from various real instruction dialogues serving as rules.Afterward, different possible strategies are applied to the newly given dialogue scenario deductively to pose various instructions.Experimental results show that our method can generate diverse and indepth instructions.The constructed multi-turn instructional dialogues can outperform competitive baselines on the downstream chat model 1 .

Recently, the remarkable advance of the Large Language Model (LLM) has inspired researchers to transfer its extraordinary reasoning capability to both vision and language data. However, the prevailing approaches primarily regard the visual input as a prompt and focus exclusively on optimizing the text generation process conditioned upon vision content by a frozen LLM. Such an inequitable treatment of vision and language heavily constrains the model's potential. In this paper, we break through this limitation by representing both vision and language in a unified form. Specifically, we introduce a well-designed visual tokenizer to translate the non-linguistic image into a sequence of discrete tokens like a foreign language that LLM can read. The resulting visual tokens encompass high-level semantics worthy of a word and also support dynamic sequence length varying from the image. Coped with this tokenizer, the presented foundation model called LaVIT can handle both image and text indiscriminately under the same generative learning paradigm. This unification empowers LaVIT to serve as an impressive generalist interface to understand and generate multi-modal content simultaneously. Extensive experiments further showcase that it outperforms the existing models by a large margin on massive vision-language tasks. Our code and models are available at https://github.com/jy0205/LaVIT.

In light of recent advances in multimodal Large Language Models (LLMs), there is increasing attention to scaling them from image-text data to more informative real-world videos. Compared to static images, video poses unique challenges for effective large-scale pre-training due to the modeling of its spatiotemporal dynamics. In this paper, we address such limitations in video-language pre-training with an efficient video decomposition that represents each video as keyframes and temporal motions. These are then adapted to an LLM using well-designed tokenizers that discretize visual and temporal information as a few tokens, thus enabling unified generative pre-training of videos, images, and text. At inference, the generated tokens from the LLM are carefully recovered to the original continuous pixel space to create various video content. Our proposed framework is both capable of comprehending and generating image and video content, as demonstrated by its competitive performance across 13 multimodal benchmarks in image and video understanding and generation. Our code and models are available at https://video-lavit.github.io.

Recently, text-to-3D generation has attracted significant attention, resulting in notable performance enhancements. Previous methods utilize end-to-end 3D generation models to initialize 3D Gaussians, multi-view diffusion models to enforce multi-view consistency, and text-to-image diffusion models to refine details with score distillation algorithms. However, these methods exhibit two limitations. Firstly, they encounter conflicts in generation directions since different models aim to produce diverse 3D assets. Secondly, the issue of over-saturation in score distillation has not been thoroughly investigated and solved. To address these limitations, we propose PlacidDreamer, a text-to-3D framework that harmonizes initialization, multi-view generation, and text-conditioned generation with a single multi-view diffusion model, while simultaneously employing a novel score distillation algorithm to achieve balanced saturation. To unify the generation direction, we introduce the Latent-Plane module, a training-friendly plug-in extension that enables multi-view diffusion models to provide fast geometry reconstruction for initialization and enhanced multi-view images to personalize the text-to-image diffusion model. To address the over-saturation problem, we propose to view score distillation as a multi-objective optimization problem and introduce the Balanced Score Distillation algorithm, which offers a Pareto Optimal solution that achieves both rich details and balanced saturation. Extensive experiments validate the outstanding capabilities of our PlacidDreamer. The code is available at https://github.com/HansenHuang0823/PlacidDreamer.

Jiao Ou, Junda Lu, Che Liu, Yihong Tang, Fuzheng Zhang, Di Zhang, Kun Gai. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Diffusion models have demonstrated great success in text-to-video (T2V) generation. However, existing methods may face challenges when handling complex (long) video generation scenarios that involve multiple objects or dynamic changes in object numbers. To address these limitations, we propose VideoTetris, a novel framework that enables compositional T2V generation. Specifically, we propose spatio-temporal compositional diffusion to precisely follow complex textual semantics by manipulating and composing the attention maps of denoising networks spatially and temporally. Moreover, we propose a new dynamic-aware data processing pipeline and a consistency regularization method to enhance the consistency of auto-regressive video generation. Extensive experiments demonstrate that our VideoTetris achieves impressive qualitative and quantitative results in compositional T2V generation. Code is available at: https://github.com/YangLing0818/VideoTetris

Brain-inspired Spiking Neural Networks (SNNs) have the characteristics of event-driven and high energy-efficient, which are different from traditional Artificial Neural Networks (ANNs) when deployed on edge devices such as neuromorphic chips. Most previous work focuses on SNNs training strategies to improve model performance and brings larger and deeper network architectures. It's difficult to deploy these complex networks on resource-limited edge devices directly. To meet such demand, people compress SNNs very cautiously to balance the performance and the computation efficiency. Existing compression methods either iteratively pruned SNNs using weights norm magnitude or formulated the problem as a sparse learning optimization. We propose an improved end-to-end Minimax optimization method for this sparse learning problem to better balance the model performance and the computation efficiency. We also demonstrate that jointly applying compression and finetuning on SNNs is better than sequentially, especially for extreme compression ratios. The compressed SNN models achieved state-of-the-art (SOTA) performance on various benchmark datasets and architectures. Our code is available athttps://github.com/chenjallen/Resource-Constrained-Compression-on-SNN .

Graph embedding based retrieval has become one of the most popular techniques in the information retrieval community and search engine industry. The classical paradigm mainly relies on the flat Euclidean geometry. In recent years, hyperbolic (negative curvature) and spherical (positive curvature) representation methods have shown their superiority to capture hierarchical and cyclic data structures respectively. However, in industrial scenarios such as e-commerce sponsored search platforms, the large-scale heterogeneous query-item-advertisement interaction graphs often have multiple structures coexisting. Existing methods either only consider a single geometry space, or combine several spaces manually, which are incapable and inflexible to model the complexity and heterogeneity in the real scenario. To tackle this challenge, we present a web-scale Adaptive Mixed-Curvature ADvertisement retrieval system (AM-CAD) to automatically capture the complex and heterogeneous graph structures in non-Euclidean spaces. Specifically, entities are represented in adaptive mixed-curvature spaces, where the types and curvatures of the subspaces are trained to be optimal combinations. Besides, an attentive edge-wise space projector is designed to model the similarities between heterogeneous nodes according to local graph structures and the relation types. Moreover, to deploy AMCAD in Taobao, one of the largest e-commerce platforms with hundreds of million users, we design an efficient two-layer online retrieval framework for the task of graph based advertisement retrieval. Extensive evaluations on real-world datasets and A/B tests on online traffic are conducted to illustrate the effectiveness of the proposed system.

The development of personalized recommendation has significantly improved the accuracy of information matching and the revenue of e-commerce platforms. Recently, it has two trends: 1) recommender systems must be trained timely to cope with ever-growing new products and ever-changing user interests from online marketing and social network; 2) state-of-the-art recommendation models introduce deep neural network (DNN) modules to improve prediction accuracy. Traditional CPU-based recommender systems cannot meet these two trends, and GPU-centric training has become a trending approach. However, we observe that GPU devices in training recommender systems are underutilized, and they cannot attain an expected throughput improvement as what it has achieved in Computer Vision (CV) and Neural Language Processing (NLP) areas. This issue can be explained by two characteristics of these recommendation models: First, they contain up to a thousand of input feature fields, introducing fragmentary and memory-intensive operations; Second, the multiple constituent feature interaction submodules introduce substantial small-sized compute kernels. To remove this roadblock to the development of recommender systems, we propose a novel framework named PICASSO to accelerate the training of recommendation models on commodity hardware. Specifically, we conduct a systematic analysis to reveal the bottlenecks encountered in training recommendation models. We leverage the model structure and data distribution to unleash the potential of hardware through our packing, interleaving, and caching optimization. Experiments show that PICASSO increases the hardware utilization by an order of magnitude on the basis of state-of-the-art baselines and brings up to 6× throughput improvement for a variety of industrial recommendation models. Using the same hardware budget in production, PICASSO on average shortens the walltime of daily training tasks by 7 hours, significantly reducing the delay of continuous delivery.

Ad retrieval in sponsored search aims to understand user search intentions (user queries) and retrieves a set of ads inferred as being relevant to the queries. Due to the huge amount of search traffic and multiple views of relevance (such as co-clicking, co-bidding or textual similar), it is highly desirable but remain challenging to achieve a large-scale, multi-view matching between queries and ads, particularly in industrial settings. In this paper, we propose a scalable multi-view ad retrieval engine SMAD that we developed and deployed at Taobao, the largest e-commerce platform in China. We construct a multi-relation query-item-ad graph capturing different views of query-ad relevance, which is of large scale and with complex structure. Since in e-commerce platform, the queries and products are organized into a category tree, to deal with the large scale of the graph, we propose a category constrained graph sampling and partition method to enable distributed parallel offline training. To tackle the complex multi-view structure, we propose a multi-view parallel deep neural network (DNN) model to combine the information from different views in a principled way. According to offline experiments and online A/B tests, our framework significantly outperforms baselines in terms of relevance, coverage, and revenue.