MoVE-KD: Knowledge Distillation for VLMs with Mixture of Visual Encoders

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 (MoVE-KD), 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 visual 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. The code will be released.

Adaptive Rate Control for Deep Video Compression with Rate-Distortion Prediction

Deep video compression has made significant progress in recent years, achieving rate-distortion performance that surpasses that of traditional video compression methods. However, rate control schemes tailored for deep video compression have not been well studied. In this paper, we propose a neural network-based $\lambda$-domain rate control scheme for deep video compression, which determines the coding parameter $\lambda$ for each to-be-coded frame based on the rate-distortion-$\lambda$ (R-D-$\lambda$) relationships directly learned from uncompressed frames, achieving high rate control accuracy efficiently without the need for pre-encoding. Moreover, this content-aware scheme is able to mitigate inter-frame quality fluctuations and adapt to abrupt changes in video content. Specifically, we introduce two neural network-based predictors to estimate the relationship between bitrate and $\lambda$, as well as the relationship between distortion and $\lambda$ for each frame. Then we determine the coding parameter $\lambda$ for each frame to achieve the target bitrate. Experimental results demonstrate that our approach achieves high rate control accuracy at the mini-GOP level with low time overhead and mitigates inter-frame quality fluctuations across video content of varying resolutions.

GraphAvatar: Compact Head Avatars with GNN-Generated 3D Gaussians

Rendering photorealistic head avatars from arbitrary viewpoints is crucial for various applications like virtual reality. Although previous methods based on Neural Radiance Fields (NeRF) can achieve impressive results, they lack fidelity and efficiency. Recent methods using 3D Gaussian Splatting (3DGS) have improved rendering quality and real-time performance but still require significant storage overhead. In this paper, we introduce a method called GraphAvatar that utilizes Graph Neural Networks (GNN) to generate 3D Gaussians for the head avatar. Specifically, GraphAvatar trains a geometric GNN and an appearance GNN to generate the attributes of the 3D Gaussians from the tracked mesh. Therefore, our method can store the GNN models instead of the 3D Gaussians, significantly reducing the storage overhead to just 10MB. To reduce the impact of face-tracking errors, we also present a novel graph-guided optimization module to refine face-tracking parameters during training. Finally, we introduce a 3D-aware enhancer for post-processing to enhance the rendering quality. We conduct comprehensive experiments to demonstrate the advantages of GraphAvatar, surpassing existing methods in visual fidelity and storage consumption. The ablation study sheds light on the trade-offs between rendering quality and model size. The code will be released at: https://github.com/ucwxb/GraphAvatar

ASGDiffusion: Parallel High-Resolution Generation with Asynchronous Structure Guidance

Training-free high-resolution (HR) image generation has garnered significant attention due to the high costs of training large diffusion models. Most existing methods begin by reconstructing the overall structure and then proceed to refine the local details. Despite their advancements, they still face issues with repetitive patterns in HR image generation. Besides, HR generation with diffusion models incurs significant computational costs. Thus, parallel generation is essential for interactive applications. To solve the above limitations, we introduce a novel method named ASGDiffusion for parallel HR generation with Asynchronous Structure Guidance (ASG) using pre-trained diffusion models. To solve the pattern repetition problem of HR image generation, ASGDiffusion leverages the low-resolution (LR) noise weighted by the attention mask as the structure guidance for the denoising step to ensure semantic consistency. The proposed structure guidance can significantly alleviate the pattern repetition problem. To enable parallel generation, we further propose a parallelism strategy, which calculates the patch noises and structure guidance asynchronously. By leveraging multi-GPU parallel acceleration, we significantly accelerate generation speed and reduce memory usage per GPU. Extensive experiments demonstrate that our method effectively and efficiently addresses common issues like pattern repetition and achieves state-of-the-art HR generation.

MixedGaussianAvatar: Realistically and Geometrically Accurate Head Avatar via Mixed 2D-3D Gaussian Splatting

Reconstructing high-fidelity 3D head avatars is crucial in various applications such as virtual reality. The pioneering methods reconstruct realistic head avatars with Neural Radiance Fields (NeRF), which have been limited by training and rendering speed. Recent methods based on 3D Gaussian Splatting (3DGS) significantly improve the efficiency of training and rendering. However, the surface inconsistency of 3DGS results in subpar geometric accuracy; later, 2DGS uses 2D surfels to enhance geometric accuracy at the expense of rendering fidelity. To leverage the benefits of both 2DGS and 3DGS, we propose a novel method named MixedGaussianAvatar for realistically and geometrically accurate head avatar reconstruction. Our main idea is to utilize 2D Gaussians to reconstruct the surface of the 3D head, ensuring geometric accuracy. We attach the 2D Gaussians to the triangular mesh of the FLAME model and connect additional 3D Gaussians to those 2D Gaussians where the rendering quality of 2DGS is inadequate, creating a mixed 2D-3D Gaussian representation. These 2D-3D Gaussians can then be animated using FLAME parameters. We further introduce a progressive training strategy that first trains the 2D Gaussians and then fine-tunes the mixed 2D-3D Gaussians. We demonstrate the superiority of MixedGaussianAvatar through comprehensive experiments. The code will be released at: https://github.com/ChenVoid/MGA/.

[CLS] Attention is All You Need for Training-Free Visual Token Pruning: Make VLM Inference Faster

Large vision-language models (VLMs) often rely on a substantial number of visual tokens when interacting with large language models (LLMs), which has proven to be inefficient. Recent efforts have aimed to accelerate VLM inference by pruning visual tokens. Most existing methods assess the importance of visual tokens based on the text-visual cross-attentions in LLMs. In this study, we find that the cross-attentions between text and visual tokens in LLMs are inaccurate. Pruning tokens based on these inaccurate attentions leads to significant performance degradation, especially at high reduction ratios. To this end, we introduce FasterVLM, a simple yet effective training-free visual token pruning method that evaluates the importance of visual tokens more accurately by utilizing attentions between the [CLS] token and image tokens from the visual encoder. Since FasterVLM eliminates redundant visual tokens immediately after the visual encoder, ensuring they do not interact with LLMs and resulting in faster VLM inference. It is worth noting that, benefiting from the accuracy of [CLS] cross-attentions, FasterVLM can prune 95\% of visual tokens while maintaining 90\% of the performance of LLaVA-1.5-7B. We apply FasterVLM to various VLMs, including LLaVA-1.5, LLaVA-NeXT, and Video-LLaVA, to demonstrate its effectiveness. Experimental results show that our FasterVLM maintains strong performance across various VLM architectures and reduction ratios, significantly outperforming existing text-visual attention-based methods. Our code is available at https://github.com/Theia-4869/FasterVLM.

EMD: Explicit Motion Modeling for High-Quality Street Gaussian Splatting

Photorealistic reconstruction of street scenes is essential for developing real-world simulators in autonomous driving. While recent methods based on 3D/4D Gaussian Splatting (GS) have demonstrated promising results, they still encounter challenges in complex street scenes due to the unpredictable motion of dynamic objects. Current methods typically decompose street scenes into static and dynamic objects, learning the Gaussians in either a supervised manner (e.g., w/ 3D bounding-box) or a self-supervised manner (e.g., w/o 3D bounding-box). However, these approaches do not effectively model the motions of dynamic objects (e.g., the motion speed of pedestrians is clearly different from that of vehicles), resulting in suboptimal scene decomposition. To address this, we propose Explicit Motion Decomposition (EMD), which models the motions of dynamic objects by introducing learnable motion embeddings to the Gaussians, enhancing the decomposition in street scenes. The proposed EMD is a plug-and-play approach applicable to various baseline methods. We also propose tailored training strategies to apply EMD to both supervised and self-supervised baselines. Through comprehensive experimentation, we illustrate the effectiveness of our approach with various established baselines. The code will be released at: https://qingpowuwu.github.io/emdgaussian.github.io/.

GazeGaussian: High-Fidelity Gaze Redirection with 3D Gaussian Splatting

Gaze estimation encounters generalization challenges when dealing with out-of-distribution data. To address this problem, recent methods use neural radiance fields (NeRF) to generate augmented data. However, existing methods based on NeRF are computationally expensive and lack facial details. 3D Gaussian Splatting (3DGS) has become the prevailing representation of neural fields. While 3DGS has been extensively examined in head avatars, it faces challenges with accurate gaze control and generalization across different subjects. In this work, we propose GazeGaussian, a high-fidelity gaze redirection method that uses a two-stream 3DGS model to represent the face and eye regions separately. By leveraging the unstructured nature of 3DGS, we develop a novel eye representation for rigid eye rotation based on the target gaze direction. To enhance synthesis generalization across various subjects, we integrate an expression-conditional module to guide the neural renderer. Comprehensive experiments show that GazeGaussian outperforms existing methods in rendering speed, gaze redirection accuracy, and facial synthesis across multiple datasets. We also demonstrate that existing gaze estimation methods can leverage GazeGaussian to improve their generalization performance. The code will be available at: https://ucwxb.github.io/GazeGaussian/.

MC-LLaVA: Multi-Concept Personalized Vision-Language Model

Current vision-language models (VLMs) show exceptional abilities across diverse tasks including visual question answering. To enhance user experience in practical applications, recent studies investigate VLM personalization to understand user-provided concepts. However, existing studies mainly focus on single-concept personalization, neglecting the existence and interplay of multiple concepts, which limits the real-world applicability of personalized VLMs. In this paper, we propose the first multi-concept personalization method named MC-LLaVA along with a high-quality multi-concept personalization dataset. Specifically, MC-LLaVA uses a joint training strategy incorporating multiple concepts in a single training step, allowing VLMs to perform accurately in multi-concept personalization. To reduce the cost of joint training, MC-LLaVA leverages visual token information for concept token initialization, yielding improved concept representation and accelerating joint training. To advance multi-concept personalization research, we further contribute a high-quality dataset. We carefully collect images from various movies that contain multiple characters and manually generate the multi-concept question-answer samples. Our dataset features diverse movie types and question-answer types. We conduct comprehensive qualitative and quantitative experiments to demonstrate that MC-LLaVA can achieve impressive multi-concept personalized responses, paving the way for VLMs to become better user-specific assistants. The code and dataset will be publicly available at https://github.com/arctanxarc/MC-LLaVA.

PLGS: Robust Panoptic Lifting with 3D Gaussian Splatting

Previous methods utilize the Neural Radiance Field (NeRF) for panoptic lifting, while their training and rendering speed are unsatisfactory. In contrast, 3D Gaussian Splatting (3DGS) has emerged as a prominent technique due to its rapid training and rendering speed. However, unlike NeRF, the conventional 3DGS may not satisfy the basic smoothness assumption as it does not rely on any parameterized structures to render (e.g., MLPs). Consequently, the conventional 3DGS is, in nature, more susceptible to noisy 2D mask supervision. In this paper, we propose a new method called PLGS that enables 3DGS to generate consistent panoptic segmentation masks from noisy 2D segmentation masks while maintaining superior efficiency compared to NeRF-based methods. Specifically, we build a panoptic-aware structured 3D Gaussian model to introduce smoothness and design effective noise reduction strategies. For the semantic field, instead of initialization with structure from motion, we construct reliable semantic anchor points to initialize the 3D Gaussians. We then use these anchor points as smooth regularization during training. Additionally, we present a self-training approach using pseudo labels generated by merging the rendered masks with the noisy masks to enhance the robustness of PLGS. For the instance field, we project the 2D instance masks into 3D space and match them with oriented bounding boxes to generate cross-view consistent instance masks for supervision. Experiments on various benchmarks demonstrate that our method outperforms previous state-of-the-art methods in terms of both segmentation quality and speed.