Vision-Language-Action (VLA) models enable robotic systems to perform embodied tasks but face deployment challenges due to the high computational demands of the dense Large Language Models (LLMs), with existing early-exit-based sparsification methods often overlooking the critical semantic role of final layers in downstream tasks. Aligning with the recent breakthrough of the Shallow Brain Hypothesis (SBH) in neuroscience and the mixture of experts in model sparsification, we conceptualize each LLM layer as an expert and propose a Mixture-of-LayEr Vision Language Action model (MoLe-VLA or simply MoLe) architecture for dynamic LLM layer activation. Specifically, we introduce a Spatial-Temporal Aware Router (STAR) for MoLe to selectively activate only parts of the layers based on the robot’s current state, mimicking the brain's distinct signal pathways specialized for cognition and causal reasoning. Additionally, to compensate for the cognition ability of LLM lost during the layer-skipping, we devise a Cognitive self-Knowledge Distillation (CogKD) to enhance the understanding of task demands and generate task-relevant action sequences by leveraging cognition features. Extensive experiments in RLBench simulations and real-world environments demonstrate the superiority of MoLe-VLA in both efficiency and performance, improving the mean success rate by 9.7% across ten simulation tasks while accelerating inference by 36.8% over OpenVLA.

Visual encoders are fundamental components in vision-language models (VLMs), each showcasing unique strengths derived from various pre-trained visual foundation models. To leverage the various capabilities of these encoders, recent studies incorporate multiple encoders within a single VLM, leading to a considerable increase in computational cost. In this paper, we present Mixture-of-Visual-Encoder Knowledge Distillation (MoVEKD), a novel framework that distills the unique proficiencies of multiple vision encoders into a single, efficient encoder model. Specifically, to mitigate conflicts and retain the unique characteristics of each teacher encoder, we employ low-rank adaptation (LoRA) and mixture-of-experts (MoEs) to selectively activate specialized knowledge based on input features, enhancing both adaptability and efficiency. To regularize the KD process and enhance performance, we propose an attention-based distillation strategy that adaptively weighs the different encoders and emphasizes valuable visual tokens, reducing the burden of replicating comprehensive but distinct features from multiple teachers. Comprehensive experiments on popular VLMs, such as LLaVA and LLaVA-NeXT, validate the effectiveness of our method. Our code is available at: https://github.com/hey-cjj/MoVE-KD.

Recent advancements in Multimodal Large Language Models (MLLMs) have shown remarkable capabilities across various multimodal contexts. However, their application in robotic scenarios, particularly for long-horizon manipulation tasks, reveals significant limitations. These limitations arise from the current MLLMs lacking three essential robotic brain capabilities: Planning Capability, which involves decomposing complex manipulation instructions into manageable sub-tasks; Affordance Perception, the ability to recognize and interpret the affordances of interactive objects; and Trajectory Prediction, the foresight to anticipate the complete manipulation trajectory necessary for successful execution. To enhance the robotic brain’s core capabilities from abstract to concrete, we introduce ShareRobot, a high-quality heterogeneous dataset that labels multi-dimensional information such as task planning, object affordance, and end-effector trajectory. ShareRobot’s diversity and accuracy have been meticulously refined by three human annotators. Building on this dataset, we developed RoboBrain, an MLLM-based model that combines robotic and general multi-modal data, utilizes a multi-stage training strategy, and incorporates long videos and high-resolution images to improve its robotic manipulation capabilities. Extensive experiments demonstrate that RoboBrain achieves state-of-the-art performance across various robotic tasks, highlighting its potential to advance robotic brain capabilities. Project website: RoboBrain.

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.

In vision-language models (VLMs), visual tokens usually consume a significant amount of computational overhead, despite their sparser information density compared to text tokens. To address this, most existing methods learn a network to prune redundant visual tokens and require additional training data. Differently, we propose an efficient training-free token optimization mechanism dubbed SparseVLM without extra parameters or fine-tuning costs. Concretely, given that visual tokens complement text tokens in VLMs for linguistic reasoning, we select visual-relevant text tokens to rate the significance of vision tokens within the self-attention matrix extracted from the VLMs. Then we progressively prune irrelevant tokens. To maximize sparsity while retaining essential information, we introduce a rank-based strategy to adaptively determine the sparsification ratio for each layer, alongside a token recycling method that compresses pruned tokens into more compact representations. Experimental results show that our SparseVLM improves the efficiency of various VLMs across a range of image and video understanding tasks. In particular, when LLaVA is equipped with SparseVLM, it achieves a 54% reduction in FLOPs, lowers CUDA time by 37%, and maintains an accuracy rate of 97%. Our code is available at https://github.com/Gumpest/SparseVLMs.

Advances in Parameter-efficient Fine-tuning (PEFT) bridged the performance gap with Full Fine-Tuning (FFT) through sophisticated analysis of pre-trained parameter spaces. Starting from drawing insights from Neural Engrams (NE) in Biological Neural Networks (BNNs), we establish a connection between the low-rank property observed during PEFT’s parameter space shifting and neurobiological mechanisms. This observation leads to our proposed method, Synapse and Neuron (SAN), which decomposes and propagates the scaling component from anterior feature adjustment vectors towards posterior weight matrices. Our approach is theoretically grounded in Long-Term Potentiation/Depression (LTP/D) phenomena, which govern synapse development through neurotransmitter release modulation. Extensive experiments demonstrate its effectiveness: on vision tasks across VTAB, FGVC, and GIC (25 datasets) using ViT, Swin-T and ConvNeXt architectures, SAN outperforms FFT up to 8.7% and LoRA by 3.2%; on language tasks using Commonsense Reasoning (8 datasets) with LLaMA models (all generations), surpassing ChatGPT up to 8.5% and LoRA by 4.7%; on vision-language tasks using Visual Instruction Tuning (7 datasets) with LLaVA models, it exceeds FFT up to 2.4% and LoRA by 1.9%. Our code and W&B log will be released

Deep learning has emerged as a pivotal tool for accelerating research in the life sciences, with the low-level processing of biomedical images (e.g., registration, fusion, restoration, super-resolution) being one of its most critical applications. Platforms such as ImageJ (Fiji) and napari have enabled the development of customized plugins for various models. However, these plugins are typically based on models that are limited to specific tasks and datasets, making them less practical for biologists. To address this challenge, we introduce **Orochi**, the first application-oriented, efficient, and versatile image processor designed to overcome these limitations. Orochi is pre-trained on patches/volumes extracted from the raw data of over 100 publicly available studies using our Random Multi-scale Sampling strategy. We further propose Task-related Joint-embedding Pre-Training (TJP), which employs biomedical task-related degradation for self-supervision rather than relying on Masked Image Modelling (MIM), which performs poorly in downstream tasks such as registration. To ensure computational efficiency, we leverage Mamba's linear computational complexity and construct Multi-head Hierarchy Mamba. Additionally, we provide a three-tier fine-tuning framework (Full, Normal, and Light) and demonstrate that Orochi achieves comparable or superior performance to current state-of-the-art specialist models, even with lightweight parameter-efficient options. We hope that our study contributes to the development of an all-in-one workflow, thereby relieving biologists from the overwhelming task of selecting among numerous models. Our pre-trained weights and code will be released.

In multimodal large language models (MLLMs), the length of input visual tokens is often significantly greater than that of their textual counterparts, leading to a high inference cost. Many works aim to address this issue by removing redundant visual tokens. However, current approaches either rely on attention-based pruning, which retains numerous duplicate tokens, or use similarity-based pruning, overlooking the instruction relevance, consequently causing suboptimal performance. In this paper, we go beyond attention or similarity by proposing a novel visual token pruning method named **CDPruner**, which maximizes the conditional diversity of retained tokens. We first define the conditional similarity between visual tokens conditioned on the instruction, and then reformulate the token pruning problem with determinantal point process (DPP) to maximize the conditional diversity of the selected subset. The proposed CDPruner is training-free and model-agnostic, allowing easy application to various MLLMs. Extensive experiments across diverse MLLMs show that CDPruner establishes new state-of-the-art on various vision-language benchmarks. By maximizing conditional diversity through DPP, the selected subset better represents the input images while closely adhering to user instructions, thereby preserving strong performance even with high reduction ratios. When applied to LLaVA, CDPruner reduces FLOPs by **95\%** and CUDA latency by **78\%**, while maintaining **94\%** of the original accuracy. Our code is available at https://github.com/Theia-4869/CDPruner.

We present the Decoupled VIdeo Segmentation (DVIS) framework, a novel approach for the challenging task of universal video segmentation, including video instance segmentation (VIS), video semantic segmentation (VSS), and video panoptic segmentation (VPS). Unlike previous methods that model video segmentation in an end-to-end manner, our approach decouples video segmentation into three cascaded sub-tasks: segmentation, tracking, and refinement. This decoupling design allows for simpler and more effective modeling of the spatio-temporal representations of objects, especially in complex scenes and long videos. Accordingly, we introduce two novel components: the referring tracker and the temporal refiner. These components track objects frame by frame and model spatio-temporal representations based on pre-aligned features. To improve the tracking capability of DVIS, we propose a denoising training strategy and introduce contrastive learning, resulting in a more robust framework named DVIS++. The proposed decoupled framework efficiently handles universal and open-vocabulary object representations, allowing DVIS++ to conduct universal and open-vocabulary video segmentation. We conduct extensive experiments on six mainstream benchmarks, including the VIS, VSS, and VPS datasets. Using a unified architecture, DVIS++ significantly outperforms state-of-the-art specialized methods on these benchmarks in closed- and open-vocabulary settings.

Animatable and relightable 3D facial generation has fundamental applications in computer vision and graphics. Although animation and relighting are highly correlated, previous methods usually address them separately. Effectively combining animation methods and relighting methods is nontrivial. In terms of explicit shading models, animatable methods cannot be easily extended to achieve realistic relighting results, such as shadow effects, due to prohibitive computational training costs. Regarding implicit lighting representations, current animatable methods cannot be incorporated due to their inharmonious animation representations, i.e., deforming spatial points. This paper, armed with a lightweight but effective lighting representation, presents a compatible animation representation to achieve a disentangled generative model of 3D animatable and relightable heads. Our represented animation allows for updating and control of realistic lighting effects. Due to the disentangled nature of our representations, we learn the animation and relighting from large-scale, in-the-wild videos instead of relying on a morphable model. We show that our method can synthesize geometrically consistent and detailed motion along with the disentangled control of lighting conditions. We further show that our method is still compatible with morphable models for driving generated avatars. Our method can also be extended to domains without video data by domain transfer to achieve a broader range of animatable and relightable head synthesis. We will release the code for reproducibility and facilitating future research.

Combining sparse IMUs and a monocular camera is a new promising setting to perform real-time human motion capture. This paper proposes a diffusion-based solution to learn human motion priors and fuse the two modalities of signals together seamlessly in a unified framework. By delicately considering the characteristics of the two signals, the sequential visual information is considered as a whole and transformed into a condition embedding, while the inertial measurement is concatenated with the noisy body pose frame by frame to construct a sequential input for the diffusion model. Firstly, we observe that the visual information may be unavailable in some frames due to occlusions or subjects moving out of the camera view. Thus incorporating the sequential visual features as a whole to get a single feature embedding is robust to the occasional degenerations of visual information in those frames. On the other hand, the IMU measurements are robust to occlusions and always stable when signal transmission has no problem. So incorporating them frame-wisely could better explore the temporal information for the system. Experiments have demonstrated the effectiveness of the system design and its state-of-the-art performance in pose estimation compared with the previous works. The code will be released.

The burgeoning field of Multimodal Large Language Models (MLLMs) has exhibited remarkable performance in diverse tasks such as captioning, commonsense reasoning, and visual scene understanding. However, the deployment of these large-scale MLLMs on client devices is hindered by their extensive model parameters, leading to a notable de-cline in generalization capabilities when these models are compressed for device deployment. Addressing this chal-lenge, we introduce a Cloud-Device Collaborative Contin-ual Adaptation framework, designed to enhance the performance of compressed, device-deployed MLLMs by lever-aging the robust capabilities of cloud-based, larger-scale MLLMs. Our framework is structured into three key components: a device-to-cloud uplink for efficient data transmission, cloud-based knowledge adaptation, and an optimized cloud-to-device downlink for model deployment. In the up-link phase, we employ an Uncertainty-guided Token Sam-pling (UTS) strategy to effectively filter out-of-distribution tokens, thereby reducing transmission costs and improving training efficiency. On the cloud side, we propose Adapter-based Knowledge Distillation (AKD) method to transfer refined knowledge from large-scale to compressed, pocket-size MLLMs. Furthermore, we propose a Dynamic Weight update Compression (DWC) strategy for the down-link, which adaptively selects and quantizes updated weight parameters, enhancing transmission efficiency and reducing the representational disparity between cloud and de-vice models. Extensive experiments on several multimodal benchmarks demonstrate the superiority of our proposed framework over prior Knowledge Distillation and device-cloud collaboration methods. Notably, we also validate the feasibility of our approach to real-world experiments.

Knowledge distillation (KD) has been applied to various tasks successfully, and mainstream methods typically boost the student model via spatial imitation losses. However, the consecutive downsamplings induced in the spatial domain of teacher model is a type of corruption, hindering the student from analyzing what specific information needs to be imitated, which results in accuracy degradation. To better understand the underlying pattern of corrupted feature maps, we shift our attention to the frequency domain. During frequency distillation, we encounter a new challenge: the low-frequency bands convey general but minimal context, while the high are more informative but also introduce noise. Not each pixel within the frequency bands contributes equally to the performance. To address the above problem: (1) We propose the Frequency Prompt plugged into the teacher model, absorbing the semantic frequency context during finetuning. (2) During the distillation period, a pixel-wise frequency mask is generated via Frequency Prompt, to localize those pixel of interests (PoIs) in various frequency bands. Additionally, we employ a position-aware relational frequency loss for dense prediction tasks, delivering a high-order spatial enhancement to the student model. We dub our Frequency Knowledge Distillation method as FreeKD, which determines the optimal localization and extent for the frequency distillation. Extensive experiments demonstrate that FreeKD not only outperforms spatial-based distillation methods consistently on dense prediction tasks (e.g., FreeKD brings 3.8 AP gains for RepPoints-R50 on COCO2017 and 4.55 mIoU gains for PSPNet-R18 on Cityscapes), but also conveys more robustness to the student. Notably, we also validate the generalization of our approach on large-scale vision models (e.g., DINO and SAM).

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.

Large vision-language models (LVLMs) have recently achieved rapid progress, exhibiting great perception and reasoning abilities concerning visual information. However, when faced with prompts in different sizes of solution spaces, LVLMs fail to always give consistent answers regarding the same knowledge point. This inconsistency of answers between different solution spaces is prevalent in LVLMs and erodes trust. To this end, we provide a multi-modal benchmark ConBench, to intuitively analyze how LVLMs perform when the solution space of a prompt revolves around a knowledge point. Based on the ConBench tool, we are the first to reveal the tapestry and get the following findings: (1) In the discriminate realm, the larger the solution space of the prompt, the lower the accuracy of the answers. (2) Establish the relationship between the discriminative and generative realms: the accuracy of the discriminative question type exhibits a strong positive correlation with its Consistency with the caption. (3) Compared to open-source models, closed-source models exhibit a pronounced bias advantage in terms of Consistency. Eventually, we ameliorate the consistency of LVLMs by trigger-based diagnostic refinement, indirectly improving the performance of their caption. We hope this paper will accelerate the research community in better evaluating their models and encourage future advancements in the consistency domain.

Video instance segmentation (VIS) is a critical task with diverse applications, including autonomous driving and video editing. Existing methods often underperform on complex and long videos in real world, primarily due to two factors. Firstly, offline methods are limited by the tightly-coupled modeling paradigm, which treats all frames equally and disregards the interdependencies between adjacent frames. Consequently, this leads to the introduction of excessive noise during long-term temporal alignment. Secondly, online methods suffer from inadequate utilization of temporal information. To tackle these challenges, we propose a decoupling strategy for VIS by dividing it into three independent sub-tasks: segmentation, tracking, and refinement. The efficacy of the decoupling strategy relies on two crucial elements: 1) attaining precise long-term alignment outcomes via frame-by-frame association during tracking, and 2) the effective utilization of temporal information predicated on the aforementioned accurate alignment outcomes during refinement. We introduce a novel referring tracker and temporal refiner to construct the Decoupled VIS framework (DVIS). DVIS achieves new SOTA performance in both VIS and VPS, surpassing the current SOTA methods by 7.3 AP and 9.6 VPQ on the OVIS and VIPSeg datasets, which are the most challenging and realistic benchmarks. Moreover, thanks to the decoupling strategy, the referring tracker and temporal refiner are super light-weight (only 1.69% of the segmenter FLOPs), allowing for efficient training and inference on a single GPU with 11G memory. The code is available at https://github.com/zhang-tao-whu/DVIS.

Knowledge distillation is an effective paradigm for boosting the performance of pocket-size model, especially when multiple teacher models are available, the student would break the upper limit again. However, it is not economical to train diverse teacher models for the disposable distillation. In this paper, we introduce a new concept dubbed Avatars for distillation, which are the inference ensemble models derived from the teacher. Concretely, (1) For each iteration of distillation training, various Avatars are generated by a perturbation transformation. We validate that Avatars own higher upper limit of working capacity and teaching ability, aiding the student model in learning diverse and receptive knowledge perspectives from the teacher model. (2) During the distillation, we propose an uncertainty-aware factor from the variance of statistical differences between the vanilla teacher and Avatars, to adjust Avatars' contribution on knowledge transfer adaptively. Avatar Knowledge Distillation (AKD) is fundamentally different from existing methods and refines with the innovative view of unequal training. Comprehensive experiments demonstrate the effectiveness of our Avatars mechanism, which polishes up the state-of-the-art distillation methods for dense prediction without more extra computational cost. The AKD brings at most 0.7 AP gains on COCO 2017 for Object Detection and 1.83 mIoU gains on Cityscapes for Semantic Segmentation, respectively.

The representation gap between teacher and student is an emerging topic in knowledge distillation (KD). To reduce the gap and improve the performance, current methods often resort to complicated training schemes, loss functions, and feature alignments, which are task-specific and feature-specific. In this paper, we state that the essence of these methods is to discard the noisy information and distill the valuable information in the feature, and propose a novel KD method dubbed DiffKD, to explicitly denoise and match features using diffusion models. Our approach is based on the observation that student features typically contain more noises than teacher features due to the smaller capacity of student model. To address this, we propose to denoise student features using a diffusion model trained by teacher features. This allows us to perform better distillation between the refined clean feature and teacher feature. Additionally, we introduce a light-weight diffusion model with a linear autoencoder to reduce the computation cost and an adaptive noise matching module to improve the denoising performance. Extensive experiments demonstrate that DiffKD is effective across various types of features and achieves state-of-the-art performance consistently on image classification, object detection, and semantic segmentation tasks. Code is available at https://github.com/hunto/DiffKD.

Learning subtle representation about object parts plays a vital role in fine-grained visual recognition (FGVR) field. The vision transformer (ViT) achieves promising results on computer vision due to its attention mechanism. Nonetheless, with the fixed size of patches in ViT, the class token in deep layer focuses on the global receptive field and cannot generate multi-granularity features for FGVR. To capture region attention without box annotations and compensate for ViT shortcomings in FGVR, we propose a novel method named Adaptive attention multi-scale Fusion Transformer (AFTrans). The Selective Attention Collection Module (SACM) in our approach leverages attention weights in ViT and filters them adaptively to correspond with the relative importance of input patches. The multiple scales (global and local) pipeline is supervised by our weights sharing encoder and can be easily trained end-to-end. Comprehensive experiments demonstrate that AFTrans can achieve SOTA performance on three published fine-grained benchmarks: CUB-200-2011, Stanford Dogs and iNat2017.