Abstract:The advancement of Multimodal Large Language Models (MLLMs) has enabled significant progress in multimodal understanding, expanding their capacity to analyze video content. However, existing evaluation benchmarks for MLLMs primarily focus on abstract video comprehension, lacking a detailed assessment of their ability to understand video compositions, the nuanced interpretation of how visual elements combine and interact within highly compiled video contexts. We introduce VidComposition, a new benchmark specifically designed to evaluate the video composition understanding capabilities of MLLMs using carefully curated compiled videos and cinematic-level annotations. VidComposition includes 982 videos with 1706 multiple-choice questions, covering various compositional aspects such as camera movement, angle, shot size, narrative structure, character actions and emotions, etc. Our comprehensive evaluation of 33 open-source and proprietary MLLMs reveals a significant performance gap between human and model capabilities. This highlights the limitations of current MLLMs in understanding complex, compiled video compositions and offers insights into areas for further improvement. The leaderboard and evaluation code are available at https://yunlong10.github.io/VidComposition/.
Abstract:Survival prediction is a critical task in pathology. In clinical practice, pathologists often examine multiple cases, leveraging a broader spectrum of cancer phenotypes to enhance pathological assessment. Despite significant advancements in deep learning, current solutions typically model each slide as a sample, struggling to effectively capture comparable and slide-agnostic pathological features. In this paper, we introduce GroupMIL, a novel framework inspired by the clinical practice of collective analysis, which models multiple slides as a single sample and organizes groups of patches and slides sequentially to capture cross-slide prognostic features. We also present GPAMamba, a model designed to facilitate intra- and inter-slide feature interactions, effectively capturing local micro-environmental characteristics within slide-level graphs while uncovering essential prognostic patterns across an extended patch sequence within the group framework. Furthermore, we develop a dual-head predictor that delivers comprehensive survival risk and probability assessments for each patient. Extensive empirical evaluations demonstrate that our model significantly outperforms state-of-the-art approaches across five datasets from The Cancer Genome Atlas.
Abstract:Achieving monocular camera localization within pre-built LiDAR maps can bypass the simultaneous mapping process of visual SLAM systems, potentially reducing the computational overhead of autonomous localization. To this end, one of the key challenges is cross-modal place recognition, which involves retrieving 3D scenes (point clouds) from a LiDAR map according to online RGB images. In this paper, we introduce an efficient framework to learn descriptors for both RGB images and point clouds. It takes visual state space model (VMamba) as the backbone and employs a pixel-view-scene joint training strategy for cross-modal contrastive learning. To address the field-of-view differences, independent descriptors are generated from multiple evenly distributed viewpoints for point clouds. A visible 3D points overlap strategy is then designed to quantify the similarity between point cloud views and RGB images for multi-view supervision. Additionally, when generating descriptors from pixel-level features using NetVLAD, we compensate for the loss of geometric information, and introduce an efficient scheme for multi-view generation. Experimental results on the KITTI and KITTI-360 datasets demonstrate the effectiveness and generalization of our method. The code will be released upon acceptance.
Abstract:Image-to-point cloud registration aims to determine the relative camera pose of an RGB image with respect to a point cloud. It plays an important role in camera localization within pre-built LiDAR maps. Despite the modality gaps, most learning-based methods establish 2D-3D point correspondences in feature space without any feedback mechanism for iterative optimization, resulting in poor accuracy and interpretability. In this paper, we propose to reformulate the registration procedure as an iterative Markov decision process, allowing for incremental adjustments to the camera pose based on each intermediate state. To achieve this, we employ reinforcement learning to develop a cross-modal registration agent (CMR-Agent), and use imitation learning to initialize its registration policy for stability and quick-start of the training. According to the cross-modal observations, we propose a 2D-3D hybrid state representation that fully exploits the fine-grained features of RGB images while reducing the useless neutral states caused by the spatial truncation of camera frustum. Additionally, the overall framework is well-designed to efficiently reuse one-shot cross-modal embeddings, avoiding repetitive and time-consuming feature extraction. Extensive experiments on the KITTI-Odometry and NuScenes datasets demonstrate that CMR-Agent achieves competitive accuracy and efficiency in registration. Once the one-shot embeddings are completed, each iteration only takes a few milliseconds.
Abstract:Image-to-point cloud registration seeks to estimate their relative camera pose, which remains an open question due to the data modality gaps. The recent matching-based methods tend to tackle this by building 2D-3D correspondences. In this paper, we reveal the information loss inherent in these methods and propose a matching-free paradigm, named MaFreeI2P. Our key insight is to actively retrieve the camera pose in SE(3) space by contrasting the geometric features between the point cloud and the query image. To achieve this, we first sample a set of candidate camera poses and construct their cost volume using the cross-modal features. Superior to matching, cost volume can preserve more information and its feature similarity implicitly reflects the confidence level of the sampled poses. Afterwards, we employ a convolutional network to adaptively formulate a similarity assessment function, where the input cost volume is further improved by filtering and pose-based weighting. Finally, we update the camera pose based on the similarity scores, and adopt a heuristic strategy to iteratively shrink the pose sampling space for convergence. Our MaFreeI2P achieves a very competitive registration accuracy and recall on the KITTI-Odometry and Apollo-DaoxiangLake datasets.
Abstract:Split federated learning (SFL) is a compute-efficient paradigm in distributed machine learning (ML), where components of large ML models are outsourced to remote servers. A significant challenge in SFL, particularly when deployed over wireless channels, is the susceptibility of transmitted model parameters to adversarial jamming that could jeopardize the learning process. This is particularly pronounced for word embedding parameters in large language models (LLMs), which are crucial for language understanding. In this paper, rigorous insights are provided into the influence of jamming LLM word embeddings in SFL by deriving an expression for the ML training loss divergence and showing that it is upper-bounded by the mean squared error (MSE). Based on this analysis, a physical layer framework is developed for resilient SFL with LLMs (R-SFLLM) over wireless networks. R-SFLLM leverages wireless sensing data to gather information on the jamming directions-of-arrival (DoAs) for the purpose of devising a novel, sensing-assisted anti-jamming strategy while jointly optimizing beamforming, user scheduling, and resource allocation. Extensive experiments using BERT and RoBERTa models demonstrate R-SFLLM's effectiveness, achieving close-to-baseline performance across various natural language processing (NLP) tasks and datasets. The proposed methodology further introduces an adversarial training component, where controlled noise exposure significantly enhances the LLM's resilience to perturbed parameters during training. The results show that more noise-sensitive models, such as RoBERTa, benefit from this feature, especially when resource allocation is unfair. It is also shown that worst-case jamming in particular translates into worst-case model outcomes, thereby necessitating the need for jamming-resilient SFL protocols.
Abstract:High-fidelity reconstruction of 3D human avatars has a wild application in visual reality. In this paper, we introduce FAGhead, a method that enables fully controllable human portraits from monocular videos. We explicit the traditional 3D morphable meshes (3DMM) and optimize the neutral 3D Gaussians to reconstruct with complex expressions. Furthermore, we employ a novel Point-based Learnable Representation Field (PLRF) with learnable Gaussian point positions to enhance reconstruction performance. Meanwhile, to effectively manage the edges of avatars, we introduced the alpha rendering to supervise the alpha value of each pixel. Extensive experimental results on the open-source datasets and our capturing datasets demonstrate that our approach is able to generate high-fidelity 3D head avatars and fully control the expression and pose of the virtual avatars, which is outperforming than existing works.
Abstract:Recent advancements in 3D generation have leveraged synthetic datasets with ground truth 3D assets and predefined cameras. However, the potential of adopting real-world datasets, which can produce significantly more realistic 3D scenes, remains largely unexplored. In this work, we delve into the key challenge of the complex and scene-specific camera trajectories found in real-world captures. We introduce Director3D, a robust open-world text-to-3D generation framework, designed to generate both real-world 3D scenes and adaptive camera trajectories. To achieve this, (1) we first utilize a Trajectory Diffusion Transformer, acting as the Cinematographer, to model the distribution of camera trajectories based on textual descriptions. (2) Next, a Gaussian-driven Multi-view Latent Diffusion Model serves as the Decorator, modeling the image sequence distribution given the camera trajectories and texts. This model, fine-tuned from a 2D diffusion model, directly generates pixel-aligned 3D Gaussians as an immediate 3D scene representation for consistent denoising. (3) Lastly, the 3D Gaussians are refined by a novel SDS++ loss as the Detailer, which incorporates the prior of the 2D diffusion model. Extensive experiments demonstrate that Director3D outperforms existing methods, offering superior performance in real-world 3D generation.
Abstract:3D open-vocabulary scene understanding, crucial for advancing augmented reality and robotic applications, involves interpreting and locating specific regions within a 3D space as directed by natural language instructions. To this end, we introduce GOI, a framework that integrates semantic features from 2D vision-language foundation models into 3D Gaussian Splatting (3DGS) and identifies 3D Gaussians of Interest using an Optimizable Semantic-space Hyperplane. Our approach includes an efficient compression method that utilizes scene priors to condense noisy high-dimensional semantic features into compact low-dimensional vectors, which are subsequently embedded in 3DGS. During the open-vocabulary querying process, we adopt a distinct approach compared to existing methods, which depend on a manually set fixed empirical threshold to select regions based on their semantic feature distance to the query text embedding. This traditional approach often lacks universal accuracy, leading to challenges in precisely identifying specific target areas. Instead, our method treats the feature selection process as a hyperplane division within the feature space, retaining only those features that are highly relevant to the query. We leverage off-the-shelf 2D Referring Expression Segmentation (RES) models to fine-tune the semantic-space hyperplane, enabling a more precise distinction between target regions and others. This fine-tuning substantially improves the accuracy of open-vocabulary queries, ensuring the precise localization of pertinent 3D Gaussians. Extensive experiments demonstrate GOI's superiority over previous state-of-the-art methods. Our project page is available at https://goi-hyperplane.github.io/ .
Abstract:We propose GGAvatar, a novel 3D avatar representation designed to robustly model dynamic head avatars with complex identities and deformations. GGAvatar employs a coarse-to-fine structure, featuring two core modules: Neutral Gaussian Initialization Module and Geometry Morph Adjuster. Neutral Gaussian Initialization Module pairs Gaussian primitives with deformable triangular meshes, employing an adaptive density control strategy to model the geometric structure of the target subject with neutral expressions. Geometry Morph Adjuster introduces deformation bases for each Gaussian in global space, creating fine-grained low-dimensional representations of deformation behaviors to address the Linear Blend Skinning formula's limitations effectively. Extensive experiments show that GGAvatar can produce high-fidelity renderings, outperforming state-of-the-art methods in visual quality and quantitative metrics.