Jiayuan Fan

Bo Zhang, Jiakang Yuan, Baopu Li et al. · deepmind

Few-shot fine-grained learning aims to classify a query image into one of a set of support categories with fine-grained differences. Although learning different objects' local differences via Deep Neural Networks has achieved success, how to exploit the query-support cross-image object semantic relations in Transformer-based architecture remains under-explored in the few-shot fine-grained scenario. In this work, we propose a Transformer-based double-helix model, namely HelixFormer, to achieve the cross-image object semantic relation mining in a bidirectional and symmetrical manner. The HelixFormer consists of two steps: 1) Relation Mining Process (RMP) across different branches, and 2) Representation Enhancement Process (REP) within each individual branch. By the designed RMP, each branch can extract fine-grained object-level Cross-image Semantic Relation Maps (CSRMs) using information from the other branch, ensuring better cross-image interaction in semantically related local object regions. Further, with the aid of CSRMs, the developed REP can strengthen the extracted features for those discovered semantically-related local regions in each branch, boosting the model's ability to distinguish subtle feature differences of fine-grained objects. Extensive experiments conducted on five public fine-grained benchmarks demonstrate that HelixFormer can effectively enhance the cross-image object semantic relation matching for recognizing fine-grained objects, achieving much better performance over most state-of-the-art methods under 1-shot and 5-shot scenarios. Our code is available at: https://github.com/JiakangYuan/HelixFormer

Peng Ye, Baopu Li, Yikang Li et al. · deepmind

Neural Architecture Search~(NAS) has attracted increasingly more attention in recent years because of its capability to design deep neural networks automatically. Among them, differential NAS approaches such as DARTS, have gained popularity for the search efficiency. However, they suffer from two main issues, the weak robustness to the performance collapse and the poor generalization ability of the searched architectures. To solve these two problems, a simple-but-efficient regularization method, termed as Beta-Decay, is proposed to regularize the DARTS-based NAS searching process. Specifically, Beta-Decay regularization can impose constraints to keep the value and variance of activated architecture parameters from too large. Furthermore, we provide in-depth theoretical analysis on how it works and why it works. Experimental results on NAS-Bench-201 show that our proposed method can help to stabilize the searching process and makes the searched network more transferable across different datasets. In addition, our search scheme shows an outstanding property of being less dependent on training time and data. Comprehensive experiments on a variety of search spaces and datasets validate the effectiveness of the proposed method.

Chongshan Lu, Fukun Yin, Xin Chen et al. · deepmind

Neural Radiance Fields (NeRF) has achieved impressive results in single object scene reconstruction and novel view synthesis, which have been demonstrated on many single modality and single object focused indoor scene datasets like DTU, BMVS, and NeRF Synthetic.However, the study of NeRF on large-scale outdoor scene reconstruction is still limited, as there is no unified outdoor scene dataset for large-scale NeRF evaluation due to expensive data acquisition and calibration costs. In this paper, we propose a large-scale outdoor multi-modal dataset, OMMO dataset, containing complex land objects and scenes with calibrated images, point clouds and prompt annotations. Meanwhile, a new benchmark for several outdoor NeRF-based tasks is established, such as novel view synthesis, surface reconstruction, and multi-modal NeRF. To create the dataset, we capture and collect a large number of real fly-view videos and select high-quality and high-resolution clips from them. Then we design a quality review module to refine images, remove low-quality frames and fail-to-calibrate scenes through a learning-based automatic evaluation plus manual review. Finally, a number of volunteers are employed to add the text descriptions for each scene and key-frame to meet the potential multi-modal requirements in the future. Compared with existing NeRF datasets, our dataset contains abundant real-world urban and natural scenes with various scales, camera trajectories, and lighting conditions. Experiments show that our dataset can benchmark most state-of-the-art NeRF methods on different tasks. We will release the dataset and model weights very soon.

Chuangguan Ye, Hongyuan Zhu, Yongbin Liao et al. · deepmind

Due to the emergence of powerful computing resources and large-scale annotated datasets, deep learning has seen wide applications in our daily life. However, most current methods require extensive data collection and retraining when dealing with novel classes never seen before. On the other hand, we humans can quickly recognize new classes by looking at a few samples, which motivates the recent popularity of few-shot learning (FSL) in machine learning communities. Most current FSL approaches work on 2D image domain, however, its implication in 3D perception is relatively under-explored. Not only needs to recognize the unseen examples as in 2D domain, 3D few-shot learning is more challenging with unordered structures, high intra-class variances, and subtle inter-class differences. Moreover, different architectures and learning algorithms make it difficult to study the effectiveness of existing 2D methods when migrating to the 3D domain. In this work, for the first time, we perform systematic and extensive studies of recent 2D FSL and 3D backbone networks for benchmarking few-shot point cloud classification, and we suggest a strong baseline and learning architectures for 3D FSL. Then, we propose a novel plug-and-play component called Cross-Instance Adaptation (CIA) module, to address the high intra-class variances and subtle inter-class differences issues, which can be easily inserted into current baselines with significant performance improvement. Extensive experiments on two newly introduced benchmark datasets, ModelNet40-FS and ShapeNet70-FS, demonstrate the superiority of our proposed network for 3D FSL.

Peng Ye, Baopu Li, Tao Chen et al. · deepmind

Semantic segmentation is a popular research topic in computer vision, and many efforts have been made on it with impressive results. In this paper, we intend to search an optimal network structure that can run in real-time for this problem. Towards this goal, we jointly search the depth, channel, dilation rate and feature spatial resolution, which results in a search space consisting of about 2.78*10^324 possible choices. To handle such a large search space, we leverage differential architecture search methods. However, the architecture parameters searched using existing differential methods need to be discretized, which causes the discretization gap between the architecture parameters found by the differential methods and their discretized version as the final solution for the architecture search. Hence, we relieve the problem of discretization gap from the innovative perspective of solution space regularization. Specifically, a novel Solution Space Regularization (SSR) loss is first proposed to effectively encourage the supernet to converge to its discrete one. Then, a new Hierarchical and Progressive Solution Space Shrinking method is presented to further achieve high efficiency of searching. In addition, we theoretically show that the optimization of SSR loss is equivalent to the L_0-norm regularization, which accounts for the improved search-evaluation gap. Comprehensive experiments show that the proposed search scheme can efficiently find an optimal network structure that yields an extremely fast speed (175 FPS) of segmentation with a small model size (1 M) while maintaining comparable accuracy.

Weimin Wu, Jiayuan Fan, Tao Chen et al. · deepmind

The linear ensemble based strategy, i.e., averaging ensemble, has been proposed to improve the performance in unsupervised domain adaptation tasks. However, a typical UDA task is usually challenged by dynamically changing factors, such as variable weather, views, and background in the unlabeled target domain. Most previous ensemble strategies ignore UDA's dynamic and uncontrollable challenge, facing limited feature representations and performance bottlenecks. To enhance the model, adaptability between domains and reduce the computational cost when deploying the ensemble model, we propose a novel framework, namely Instance aware Model Ensemble With Distillation, IMED, which fuses multiple UDA component models adaptively according to different instances and distills these components into a small model. The core idea of IMED is a dynamic instance aware ensemble strategy, where for each instance, a nonlinear fusion subnetwork is learned that fuses the extracted features and predicted labels of multiple component models. The nonlinear fusion method can help the ensemble model handle dynamically changing factors. After learning a large capacity ensemble model with good adaptability to different changing factors, we leverage the ensemble teacher model to guide the learning of a compact student model by knowledge distillation. Furthermore, we provide the theoretical analysis of the validity of IMED for UDA. Extensive experiments conducted on various UDA benchmark datasets, e.g., Office 31, Office Home, and VisDA 2017, show the superiority of the model based on IMED to the state of the art methods under the comparable computation cost.

Sijin Chen, Xin Chen, Chi Zhang et al.

Recent advances in Large Multimodal Models (LMM) have made it possible for various applications in human-machine interactions. However, developing LMMs that can comprehend, reason, and plan in complex and diverse 3D environments remains a challenging topic, especially considering the demand for understanding permutation-invariant point cloud 3D representations of the 3D scene. Existing works seek help from multi-view images, and project 2D features to 3D space as 3D scene representations. This, however, leads to huge computational overhead and performance degradation. In this paper, we present LL3DA, a Large Language 3D Assistant that takes point cloud as direct input and respond to both textual-instructions and visual-prompts. This help LMMs better comprehend human interactions and further help to remove the ambiguities in cluttered 3D scenes. Experiments show that LL3DA achieves remarkable results, and surpasses various 3D vision-language models on both 3D Dense Captioning and 3D Question Answering.

Hancheng Ye, Bo Zhang, Tao Chen et al.

Global channel pruning (GCP) aims to remove a subset of channels (filters) across different layers from a deep model without hurting the performance. Previous works focus on either single task model pruning or simply adapting it to multitask scenario, and still face the following problems when handling multitask pruning: 1) Due to the task mismatch, a well-pruned backbone for classification task focuses on preserving filters that can extract category-sensitive information, causing filters that may be useful for other tasks to be pruned during the backbone pruning stage; 2) For multitask predictions, different filters within or between layers are more closely related and interacted than that for single task prediction, making multitask pruning more difficult. Therefore, aiming at multitask model compression, we propose a Performance-Aware Global Channel Pruning (PAGCP) framework. We first theoretically present the objective for achieving superior GCP, by considering the joint saliency of filters from intra- and inter-layers. Then a sequentially greedy pruning strategy is proposed to optimize the objective, where a performance-aware oracle criterion is developed to evaluate sensitivity of filters to each task and preserve the globally most task-related filters. Experiments on several multitask datasets show that the proposed PAGCP can reduce the FLOPs and parameters by over 60% with minor performance drop, and achieves 1.2x$\sim$3.3x acceleration on both cloud and mobile platforms.

Jingyi Zhou, Jiamu Sheng, Jiayuan Fan et al.

The effectiveness of spectral-spatial feature learning is crucial for the hyperspectral image (HSI) classification task. Diffusion models, as a new class of groundbreaking generative models, have the ability to learn both contextual semantics and textual details from the distinct timestep dimension, enabling the modeling of complex spectral-spatial relations in HSIs. However, existing diffusion-based HSI classification methods only utilize manually selected single-timestep single-stage features, limiting the full exploration and exploitation of rich contextual semantics and textual information hidden in the diffusion model. To address this issue, we propose a novel diffusion-based feature learning framework that explores Multi-Timestep Multi-Stage Diffusion features for HSI classification for the first time, called MTMSD. Specifically, the diffusion model is first pretrained with unlabeled HSI patches to mine the connotation of unlabeled data, and then is used to extract the multi-timestep multi-stage diffusion features. To effectively and efficiently leverage multi-timestep multi-stage features,two strategies are further developed. One strategy is class & timestep-oriented multi-stage feature purification module with the inter-class and inter-timestep prior for reducing the redundancy of multi-stage features and alleviating memory constraints. The other one is selective timestep feature fusion module with the guidance of global features to adaptively select different timestep features for integrating texture and semantics. Both strategies facilitate the generality and adaptability of the MTMSD framework for diverse patterns of different HSI data. Extensive experiments are conducted on four public HSI datasets, and the results demonstrate that our method outperforms state-of-the-art methods for HSI classification, especially on the challenging Houston 2018 dataset.

Fukun Yin, Xin Chen, Chi Zhang et al.

The advent of large language models, enabling flexibility through instruction-driven approaches, has revolutionized many traditional generative tasks, but large models for 3D data, particularly in comprehensively handling 3D shapes with other modalities, are still under-explored. By achieving instruction-based shape generations, versatile multimodal generative shape models can significantly benefit various fields like 3D virtual construction and network-aided design. In this work, we present ShapeGPT, a shape-included multi-modal framework to leverage strong pre-trained language models to address multiple shape-relevant tasks. Specifically, ShapeGPT employs a word-sentence-paragraph framework to discretize continuous shapes into shape words, further assembles these words for shape sentences, as well as integrates shape with instructional text for multi-modal paragraphs. To learn this shape-language model, we use a three-stage training scheme, including shape representation, multimodal alignment, and instruction-based generation, to align shape-language codebooks and learn the intricate correlations among these modalities. Extensive experiments demonstrate that ShapeGPT achieves comparable performance across shape-relevant tasks, including text-to-shape, shape-to-text, shape completion, and shape editing.

Yuhan Ding, Fukun Yin, Jiayuan Fan et al.

Recent advances in implicit neural representations have achieved impressive results by sampling and fusing individual points along sampling rays in the sampling space. However, due to the explosively growing sampling space, finely representing and synthesizing detailed textures remains a challenge for unbounded large-scale outdoor scenes. To alleviate the dilemma of using individual points to perceive the entire colossal space, we explore learning the surface distribution of the scene to provide structural priors and reduce the samplable space and propose a Point Diffusion implicit Function, PDF, for large-scale scene neural representation. The core of our method is a large-scale point cloud super-resolution diffusion module that enhances the sparse point cloud reconstructed from several training images into a dense point cloud as an explicit prior. Then in the rendering stage, only sampling points with prior points within the sampling radius are retained. That is, the sampling space is reduced from the unbounded space to the scene surface. Meanwhile, to fill in the background of the scene that cannot be provided by point clouds, the region sampling based on Mip-NeRF 360 is employed to model the background representation. Expensive experiments have demonstrated the effectiveness of our method for large-scale scene novel view synthesis, which outperforms relevant state-of-the-art baselines.

Lin Zhan, Jiayuan Fan, Peng Ye et al.

Existing deep learning-based hyperspectral image (HSI) classification works still suffer from the limitation of the fixed-sized receptive field, leading to difficulties in distinctive spectral-spatial features for ground objects with various sizes and arbitrary shapes. Meanwhile, plenty of previous works ignore asymmetric spectral-spatial dimensions in HSI. To address the above issues, we propose a multi-stage search architecture in order to overcome asymmetric spectral-spatial dimensions and capture significant features. First, the asymmetric pooling on the spectral-spatial dimension maximally retains the essential features of HSI. Then, the 3D convolution with a selectable range of receptive fields overcomes the constraints of fixed-sized convolution kernels. Finally, we extend these two searchable operations to different layers of each stage to build the final architecture. Extensive experiments are conducted on two challenging HSI benchmarks including Indian Pines and Houston University, and results demonstrate the effectiveness of the proposed method with superior performance compared with the related works.

Yiying Yang, Wen Liu, Fukun Yin et al.

Recent advancements in implicit neural representations have contributed to high-fidelity surface reconstruction and photorealistic novel view synthesis. However, the computational complexity inherent in these methodologies presents a substantial impediment, constraining the attainable frame rates and resolutions in practical applications. In response to this predicament, we propose VQ-NeRF, an effective and efficient pipeline for enhancing implicit neural representations via vector quantization. The essence of our method involves reducing the sampling space of NeRF to a lower resolution and subsequently reinstating it to the original size utilizing a pre-trained VAE decoder, thereby effectively mitigating the sampling time bottleneck encountered during rendering. Although the codebook furnishes representative features, reconstructing fine texture details of the scene remains challenging due to high compression rates. To overcome this constraint, we design an innovative multi-scale NeRF sampling scheme that concurrently optimizes the NeRF model at both compressed and original scales to enhance the network's ability to preserve fine details. Furthermore, we incorporate a semantic loss function to improve the geometric fidelity and semantic coherence of our 3D reconstructions. Extensive experiments demonstrate the effectiveness of our model in achieving the optimal trade-off between rendering quality and efficiency. Evaluation on the DTU, BlendMVS, and H3DS datasets confirms the superior performance of our approach.

Jiamu Sheng, Jiayuan Fan, Peng Ye et al.

Despite substantial progress in no-reference image quality assessment (NR-IQA), previous training models often suffer from over-fitting due to the limited scale of used datasets, resulting in model performance bottlenecks. To tackle this challenge, we explore the potential of leveraging data augmentation to improve data efficiency and enhance model robustness. However, most existing data augmentation methods incur a serious issue, namely that it alters the image quality and leads to training images mismatching with their original labels. Additionally, although only a few data augmentation methods are available for NR-IQA task, their ability to enrich dataset diversity is still insufficient. To address these issues, we propose a effective and general data augmentation based on just noticeable difference (JND) noise mixing for NR-IQA task, named JNDMix. In detail, we randomly inject the JND noise, imperceptible to the human visual system (HVS), into the training image without any adjustment to its label. Extensive experiments demonstrate that JNDMix significantly improves the performance and data efficiency of various state-of-the-art NR-IQA models and the commonly used baseline models, as well as the generalization ability. More importantly, JNDMix facilitates MANIQA to achieve the state-of-the-art performance on LIVEC and KonIQ-10k.

Yiying Yang, Fukun Yin, Jiayuan Fan et al.

As Artificial Intelligence Generated Content (AIGC) advances, a variety of methods have been developed to generate text, images, videos, and 3D objects from single or multimodal inputs, contributing efforts to emulate human-like cognitive content creation. However, generating realistic large-scale scenes from a single input presents a challenge due to the complexities involved in ensuring consistency across extrapolated views generated by models. Benefiting from recent video generation models and implicit neural representations, we propose Scene123, a 3D scene generation model, that not only ensures realism and diversity through the video generation framework but also uses implicit neural fields combined with Masked Autoencoders (MAE) to effectively ensures the consistency of unseen areas across views. Specifically, we initially warp the input image (or an image generated from text) to simulate adjacent views, filling the invisible areas with the MAE model. However, these filled images usually fail to maintain view consistency, thus we utilize the produced views to optimize a neural radiance field, enhancing geometric consistency. Moreover, to further enhance the details and texture fidelity of generated views, we employ a GAN-based Loss against images derived from the input image through the video generation model. Extensive experiments demonstrate that our method can generate realistic and consistent scenes from a single prompt. Both qualitative and quantitative results indicate that our approach surpasses existing state-of-the-art methods. We show encourage video examples at https://yiyingyang12.github.io/Scene123.github.io/.

Shenghe Zheng, Hongzhi Wang, Chenyu Huang et al.

With more open-source models available for diverse tasks, model merging has gained attention by combining models into one, reducing training, storage, and inference costs. Current research mainly focuses on model merging for full fine-tuning, overlooking the popular LoRA. However, our empirical analysis reveals that: a) existing merging methods designed for full fine-tuning perform poorly on LoRA; b) LoRA modules show much larger parameter magnitude variance than full fine-tuned weights; c) greater parameter magnitude variance correlates with worse merging performance. Considering that large magnitude variances cause deviations in the distribution of the merged parameters, resulting in information loss and performance degradation, we propose a Decoupled and Orthogonal merging approach(DO-Merging). By separating parameters into magnitude and direction components and merging them independently, we reduce the impact of magnitude differences on the directional alignment of the merged models, thereby preserving task information. Furthermore, we introduce a data-free, layer-wise gradient descent method with orthogonal constraints to mitigate interference during the merging of direction components. We provide theoretical guarantees for both the decoupling and orthogonal components. And we validate through extensive experiments across vision, language, and multi-modal domains that our proposed DO-Merging can achieve significantly higher performance than existing merging methods at a minimal cost. Notably, each component can be flexibly integrated with existing methods, offering near free-lunch improvements across tasks.

Zhende Song, Shengji Tang, Peng Ye et al.

Test-time scaling (TTS) has emerged as a promising, training-free approach for enhancing large language model (LLM) performance. However, the efficacy of existing methods, such as Best-of-N and Self-Consistency, is fundamentally constrained by the dominant single test-time scaling (STTS) paradigm, which relies on a single LLM agent interacting with a single reward model (SA-SR). Inspired by recent work showing that collective methods can surpass the performance ceiling of individual models, we introduce Collective Test-Time Scaling (CTTS). First, we systematically investigate three primary interaction paradigms of existing multiple models: single-agent-multi-reward (SA-MR), multi-agent-single-reward (MA-SR), and multi-agent-multi-reward (MA-MR). Extensive experiments reveal that the MA-MR paradigm is consistently superior. Based on this finding, we further propose CTTS-MM, a novel framework that operationalizes multi-agent and multi-reward collaboration. CTTS-MM integrates two key technical contributions: (1) for agent collaboration, an Agent Collaboration Search (ACS) that identifies the most effective combination of LLMs from a candidate pool; and (2) for reward model collaboration, a Mixture of Reward Models (MoR) strategy that leverages a Prior Reward model Ensemble Selection (PRES) algorithm to select the optimal ensemble. Evaluations across seven mainstream benchmarks demonstrate that CTTS-MM significantly outperforms leading STTS methods (+4.82% over Best-of-N) and surpasses even flagship proprietary LLMs (+7.06% over GPT-4.1) and open-source LLMs. These results highlight the substantial potential of collective scaling to push the frontier of LLM inference. Code will be released at https://github.com/magent4aci/CTTS-MM.

Mingsheng Li, Lin Zhang, Mingzhen Zhu et al.

This paper studies the problem of pre-training for small models, which is essential for many mobile devices. Current state-of-the-art methods on this problem transfer the representational knowledge of a large network (as a Teacher) into a smaller model (as a Student) using self-supervised distillation, improving the performance of the small model on downstream tasks. However, existing approaches are insufficient in extracting the crucial knowledge that is useful for discerning categories in downstream tasks during the distillation process. In this paper, for the first time, we introduce language guidance to the distillation process and propose a new method named Language-Guided Distillation (LGD) system, which uses category names of the target downstream task to help refine the knowledge transferred between the teacher and student. To this end, we utilize a pre-trained text encoder to extract semantic embeddings from language and construct a textual semantic space called Textual Semantics Bank (TSB). Furthermore, we design a Language-Guided Knowledge Aggregation (LGKA) module to construct the visual semantic space, also named Visual Semantics Bank (VSB). The task-related knowledge is transferred by driving a student encoder to mimic the similarity score distribution inferred by a teacher over TSB and VSB. Compared with other small models obtained by either ImageNet pre-training or self-supervised distillation, experiment results show that the distilled lightweight model using the proposed LGD method presents state-of-the-art performance and is validated on various downstream tasks, including classification, detection, and segmentation. We have made the code available at https://github.com/mZhenz/LGD.

Jingjing Chen, Jichao Zhang, Enver Sangineto et al.

Gaze redirection aims at manipulating the gaze of a given face image with respect to a desired direction (i.e., a reference angle) and it can be applied to many real life scenarios, such as video-conferencing or taking group photos. However, previous work on this topic mainly suffers of two limitations: (1) Low-quality image generation and (2) Low redirection precision. In this paper, we propose to alleviate these problems by means of a novel gaze redirection framework which exploits both a numerical and a pictorial direction guidance, jointly with a coarse-to-fine learning strategy. Specifically, the coarse branch learns the spatial transformation which warps input image according to desired gaze. On the other hand, the fine-grained branch consists of a generator network with conditional residual image learning and a multi-task discriminator. This second branch reduces the gap between the previously warped image and the ground-truth image and recovers finer texture details. Moreover, we propose a numerical and pictorial guidance module~(NPG) which uses a pictorial gazemap description and numerical angles as an extra guide to further improve the precision of gaze redirection. Extensive experiments on a benchmark dataset show that the proposed method outperforms the state-of-the-art approaches in terms of both image quality and redirection precision. The code is available at https://github.com/jingjingchen777/CFGR

Biao Jiang, Xin Chen, Chi Zhang et al.

Recent advancements in language models have demonstrated their adeptness in conducting multi-turn dialogues and retaining conversational context. However, this proficiency remains largely unexplored in other multimodal generative models, particularly in human motion models. By integrating multi-turn conversations in controlling continuous virtual human movements, generative human motion models can achieve an intuitive and step-by-step process of human task execution for humanoid robotics, game agents, or other embodied systems. In this work, we present MotionChain, a conversational human motion controller to generate continuous and long-term human motion through multimodal prompts. Specifically, MotionChain consists of multi-modal tokenizers that transform various data types such as text, image, and motion, into discrete tokens, coupled with a Vision-Motion-aware Language model. By leveraging large-scale language, vision-language, and vision-motion data to assist motion-related generation tasks, MotionChain thus comprehends each instruction in multi-turn conversation and generates human motions followed by these prompts. Extensive experiments validate the efficacy of MotionChain, demonstrating state-of-the-art performance in conversational motion generation, as well as more intuitive manners of controlling and interacting with virtual humans.

Zhende Song, Chenchen Wang, Jiamu Sheng et al.

Recent large vision-language models (LVLMs) for video understanding are primarily fine-tuned with various videos scraped from online platforms. Existing datasets, such as ActivityNet, require considerable human labor for structuring and annotation before effectively utilized for tuning LVLMs. While current LVLMs are primarily trained on existing datasets in broad, general-purpose settings, adapting them to specific downstream scenarios remains challenging, as collecting and annotating task-specific videos is highly labor-intensive and time-consuming. To address this issue, we propose a three-stage framework named DreamFrame for automatically generating style-consistent keyframes and corresponding question-answer (QA) pairs to support LVLM instruction tuning. DreamFrame generates datasets in a movie-like manner. First, we utilize an LLM to generate structured movie plots including movie prior information (like overview and style), frame descriptions and plot-related QA pairs, with a story expansion strategy to mitigate context length limitations.Then, to ensure visual consistency across generated frames, we design a Style Immobilization Process which maintains consistent style through an embedding learning strategy. Finally, frame descriptions and style embeddings are integrated to produce coherent keyframes. Using DreamFrame, we construct a dataset comprising approximately 1k stylized keyframe-like videos and 100k diverse QA pairs. Extensive fine-tuned experiments on various LVLM architectures demonstrate the effectiveness of the proposed dataset. Furthermore, based on the proposed dataset, we fine-tune a new LVLM named DreamFrame-7B, which significantly surpasses the previous similar-sized LVLMs across different benchmarks.

Jingdong Zhang, Jiayuan Fan, Peng Ye et al.

Multi-task dense prediction aims at handling multiple pixel-wise prediction tasks within a unified network simultaneously for visual scene understanding. However, cross-task feature interactions of current methods are still suffering from incomplete levels of representations, less discriminative semantics in feature participants, and inefficient pair-wise task interaction processes. To tackle these under-explored issues, we propose a novel BridgeNet framework, which extracts comprehensive and discriminative intermediate Bridge Features, and conducts interactions based on them. Specifically, a Task Pattern Propagation (TPP) module is firstly applied to ensure highly semantic task-specific feature participants are prepared for subsequent interactions, and a Bridge Feature Extractor (BFE) is specially designed to selectively integrate both high-level and low-level representations to generate the comprehensive bridge features. Then, instead of conducting heavy pair-wise cross-task interactions, a Task-Feature Refiner (TFR) is developed to efficiently take guidance from bridge features and form final task predictions. To the best of our knowledge, this is the first work considering the completeness and quality of feature participants in cross-task interactions. Extensive experiments are conducted on NYUD-v2, Cityscapes and PASCAL Context benchmarks, and the superior performance shows the proposed architecture is effective and powerful in promoting different dense prediction tasks simultaneously.

Benjia Zhou, Bin Fu, Pei Cheng et al.

Recent advances in large-scale text-to-image diffusion models (e.g., FLUX.1) have greatly improved visual fidelity in consistent character generation and editing. However, existing methods rarely unify these tasks within a single framework. Generation-based approaches struggle with fine-grained identity consistency across instances, while editing-based methods often lose spatial controllability and instruction alignment. To bridge this gap, we propose ReMix, a unified framework for character-consistent generation and editing. It constitutes two core components: the ReMix Module and IP-ControlNet. The ReMix Module leverages the multimodal reasoning ability of MLLMs to edit semantic features of input images and adapt instruction embeddings to the native DiT backbone without fine-tuning. While this ensures coherent semantic layouts, pixel-level consistency and pose controllability remain challenging. To address this, IP-ControlNet extends ControlNet to decouple semantic and layout cues from reference images and introduces an ε-equivariant latent space that jointly denoises the reference and target images within a shared noise space. Inspired by convergent evolution and quantum decoherence,i.e., where environmental noise drives state convergence, this design promotes feature alignment in the hidden space, enabling consistent object generation while preserving identity. ReMix supports a wide range of tasks, including personalized generation, image editing, style transfer, and multi-condition synthesis. Extensive experiments validate its effectiveness and efficiency as a unified framework for character-consistent image generation and editing.

Jiamu Sheng, Jingyi Zhou, Jiong Wang et al.

The effectiveness and efficiency of modeling complex spectral-spatial relations are both crucial for Hyperspectral image (HSI) classification. Most existing methods based on CNNs and transformers still suffer from heavy computational burdens and have room for improvement in capturing the global-local spectral-spatial feature representation. To this end, we propose a novel lightweight parallel design called lightweight dual-stream Mamba-convolution network (DualMamba) for HSI classification. Specifically, a parallel lightweight Mamba and CNN block are first developed to extract global and local spectral-spatial features. First, the cross-attention spectral-spatial Mamba module is proposed to leverage the global modeling of Mamba at linear complexity. Within this module, dynamic positional embedding is designed to enhance the spatial location information of visual sequences. The lightweight spectral/spatial Mamba blocks comprise an efficient scanning strategy and a lightweight Mamba design to efficiently extract global spectral-spatial features. And the cross-attention spectral-spatial fusion is designed to learn cross-correlation and fuse spectral-spatial features. Second, the lightweight spectral-spatial residual convolution module is proposed with lightweight spectral and spatial branches to extract local spectral-spatial features through residual learning. Finally, the adaptive global-local fusion is proposed to dynamically combine global Mamba features and local convolution features for a global-local spectral-spatial representation. Compared with state-of-the-art HSI classification methods, experimental results demonstrate that DualMamba achieves significant classification accuracy on three public HSI datasets and a superior reduction in model parameters and floating point operations (FLOPs).

Minglei Li, Peng Ye, Yongqi Huang et al.

Parameter-efficient fine-tuning (PEFT) has become increasingly important as foundation models continue to grow in both popularity and size. Adapter has been particularly well-received due to their potential for parameter reduction and adaptability across diverse tasks. However, striking a balance between high efficiency and robust generalization across tasks remains a challenge for adapter-based methods. We analyze existing methods and find that: 1) parameter sharing is the key to reducing redundancy; 2) more tunable parameters, dynamic allocation, and block-specific design are keys to improving performance. Unfortunately, no previous work considers all these factors. Inspired by this insight, we introduce a novel framework named Adapter-X. First, a Sharing Mixture of Adapters (SMoA) module is proposed to fulfill token-level dynamic allocation, increased tunable parameters, and inter-block sharing at the same time. Second, some block-specific designs like Prompt Generator (PG) are introduced to further enhance the ability of adaptation. Extensive experiments across 2D image and 3D point cloud modalities demonstrate that Adapter-X represents a significant milestone as it is the first to outperform full fine-tuning in both 2D image and 3D point cloud modalities with significantly fewer parameters, i.e., only 0.20% and 1.88% of original trainable parameters for 2D and 3D classification tasks. Our code will be publicly available.

Chong Yu, Tao Chen, Zhongxue Gan et al.

The transformer extends its success from the language to the vision domain. Because of the stacked self-attention and cross-attention blocks, the acceleration deployment of vision transformer on GPU hardware is challenging and also rarely studied. This paper thoroughly designs a compression scheme to maximally utilize the GPU-friendly 2:4 fine-grained structured sparsity and quantization. Specially, an original large model with dense weight parameters is first pruned into a sparse one by 2:4 structured pruning, which considers the GPU's acceleration of 2:4 structured sparse pattern with FP16 data type, then the floating-point sparse model is further quantized into a fixed-point one by sparse-distillation-aware quantization aware training, which considers GPU can provide an extra speedup of 2:4 sparse calculation with integer tensors. A mixed-strategy knowledge distillation is used during the pruning and quantization process. The proposed compression scheme is flexible to support supervised and unsupervised learning styles. Experiment results show GPUSQ-ViT scheme achieves state-of-the-art compression by reducing vision transformer models 6.4-12.7 times on model size and 30.3-62 times on FLOPs with negligible accuracy degradation on ImageNet classification, COCO detection and ADE20K segmentation benchmarking tasks. Moreover, GPUSQ-ViT can boost actual deployment performance by 1.39-1.79 times and 3.22-3.43 times of latency and throughput on A100 GPU, and 1.57-1.69 times and 2.11-2.51 times improvement of latency and throughput on AGX Orin.

Yongbin Liao, Hongyuan Zhu, Yanggang Zhang et al.

Point cloud instance segmentation has achieved huge progress with the emergence of deep learning. However, these methods are usually data-hungry with expensive and time-consuming dense point cloud annotations. To alleviate the annotation cost, unlabeled or weakly labeled data is still less explored in the task. In this paper, we introduce the first semi-supervised point cloud instance segmentation framework (SPIB) using both labeled and unlabelled bounding boxes as supervision. To be specific, our SPIB architecture involves a two-stage learning procedure. For stage one, a bounding box proposal generation network is trained under a semi-supervised setting with perturbation consistency regularization (SPCR). The regularization works by enforcing an invariance of the bounding box predictions over different perturbations applied to the input point clouds, to provide self-supervision for network learning. For stage two, the bounding box proposals with SPCR are grouped into some subsets, and the instance masks are mined inside each subset with a novel semantic propagation module and a property consistency graph module. Moreover, we introduce a novel occupancy ratio guided refinement module to refine the instance masks. Extensive experiments on the challenging ScanNet v2 dataset demonstrate our method can achieve competitive performance compared with the recent fully-supervised methods.

Bo Zhang, Tao Chen, Bin Wang et al.

Unsupervised domain adaptive object detection aims to adapt a well-trained detector from its original source domain with rich labeled data to a new target domain with unlabeled data. Previous works focus on improving the domain adaptability of region-based detectors, e.g., Faster-RCNN, through matching cross-domain instance-level features that are explicitly extracted from a region proposal network (RPN). However, this is unsuitable for region-free detectors such as single shot detector (SSD), which perform a dense prediction from all possible locations in an image and do not have the RPN to encode such instance-level features. As a result, they fail to align important image regions and crucial instance-level features between the domains of region-free detectors. In this work, we propose an adversarial module to strengthen the cross-domain matching of instance-level features for region-free detectors. Firstly, to emphasize the important regions of image, the DSEM learns to predict a transferable foreground enhancement mask that can be utilized to suppress the background disturbance in an image. Secondly, considering that region-free detectors recognize objects of different scales using multi-scale feature maps, the DSEM encodes both multi-level semantic representations and multi-instance spatial-contextual relationships across different domains. Finally, the DSEM is pluggable into different region-free detectors, ultimately achieving the densely semantic feature matching via adversarial learning. Extensive experiments have been conducted on PASCAL VOC, Clipart, Comic, Watercolor, and FoggyCityscape benchmarks, and their results well demonstrate that the proposed approach not only improves the domain adaptability of region-free detectors but also outperforms existing domain adaptive region-based detectors under various domain shift settings.

Yike Wu, Bo Zhang, Gang Yu et al.

The goal of few-shot fine-grained image classification is to recognize rarely seen fine-grained objects in the query set, given only a few samples of this class in the support set. Previous works focus on learning discriminative image features from a limited number of training samples for distinguishing various fine-grained classes, but ignore one important fact that spatial alignment of the discriminative semantic features between the query image with arbitrary changes and the support image, is also critical for computing the semantic similarity between each support-query pair. In this work, we propose an object-aware long-short-range spatial alignment approach, which is composed of a foreground object feature enhancement (FOE) module, a long-range semantic correspondence (LSC) module and a short-range spatial manipulation (SSM) module. The FOE is developed to weaken background disturbance and encourage higher foreground object response. To address the problem of long-range object feature misalignment between support-query image pairs, the LSC is proposed to learn the transferable long-range semantic correspondence by a designed feature similarity metric. Further, the SSM module is developed to refine the transformed support feature after the long-range step to align short-range misaligned features (or local details) with the query features. Extensive experiments have been conducted on four benchmark datasets, and the results show superior performance over most state-of-the-art methods under both 1-shot and 5-shot classification scenarios.

Qihang Yang, Tao Chen, Jiayuan Fan et al.

Due to real-time image semantic segmentation needs on power constrained edge devices, there has been an increasing desire to design lightweight semantic segmentation neural network, to simultaneously reduce computational cost and increase inference speed. In this paper, we propose an efficient asymmetric dilated semantic segmentation network, named EADNet, which consists of multiple developed asymmetric convolution branches with different dilation rates to capture the variable shapes and scales information of an image. Specially, a multi-scale multi-shape receptive field convolution (MMRFC) block with only a few parameters is designed to capture such information. Experimental results on the Cityscapes dataset demonstrate that our proposed EADNet achieves segmentation mIoU of 67.1 with smallest number of parameters (only 0.35M) among mainstream lightweight semantic segmentation networks.

Jingchen Sun, Jiming Chen, Tao Chen et al.

Vision-based dynamic pedestrian intrusion detection (PID), judging whether pedestrians intrude an area-of-interest (AoI) by a moving camera, is an important task in mobile surveillance. The dynamically changing AoIs and a number of pedestrians in video frames increase the difficulty and computational complexity of determining whether pedestrians intrude the AoI, which makes previous algorithms incapable of this task. In this paper, we propose a novel and efficient multi-task deep neural network, PIDNet, to solve this problem. PIDNet is mainly designed by considering two factors: accurately segmenting the dynamically changing AoIs from a video frame captured by the moving camera and quickly detecting pedestrians from the generated AoI-contained areas. Three efficient network designs are proposed and incorporated into PIDNet to reduce the computational complexity: 1) a special PID task backbone for feature sharing, 2) a feature cropping module for feature cropping, and 3) a lighter detection branch network for feature compression. In addition, considering there are no public datasets and benchmarks in this field, we establish a benchmark dataset to evaluate the proposed network and give the corresponding evaluation metrics for the first time. Experimental results show that PIDNet can achieve 67.1% PID accuracy and 9.6 fps inference speed on the proposed dataset, which serves as a good baseline for the future vision-based dynamic PID study.