Vision-Centric Activation and Coordination for Multimodal Large Language Models

Multimodal large language models (MLLMs) integrate image features from visual encoders with LLMs, demonstrating advanced comprehension capabilities. However, mainstream MLLMs are solely supervised by the next-token prediction of textual tokens, neglecting critical vision-centric information essential for analytical abilities. To track this dilemma, we introduce VaCo, which optimizes MLLM representations through Vision-Centric activation and Coordination from multiple vision foundation models (VFMs). VaCo introduces visual discriminative alignment to integrate task-aware perceptual features extracted from VFMs, thereby unifying the optimization of both textual and visual outputs in MLLMs. Specifically, we incorporate the learnable Modular Task Queries (MTQs) and Visual Alignment Layers (VALs) into MLLMs, activating specific visual signals under the supervision of diverse VFMs. To coordinate representation conflicts across VFMs, the crafted Token Gateway Mask (TGM) restricts the information flow among multiple groups of MTQs. Extensive experiments demonstrate that VaCo significantly improves the performance of different MLLMs on various benchmarks, showcasing its superior capabilities in visual comprehension.

Scene Graph Disentanglement and Composition for Generalizable Complex Image Generation

There has been exciting progress in generating images from natural language or layout conditions. However, these methods struggle to faithfully reproduce complex scenes due to the insufficient modeling of multiple objects and their relationships. To address this issue, we leverage the scene graph, a powerful structured representation, for complex image generation. Different from the previous works that directly use scene graphs for generation, we employ the generative capabilities of variational autoencoders and diffusion models in a generalizable manner, compositing diverse disentangled visual clues from scene graphs. Specifically, we first propose a Semantics-Layout Variational AutoEncoder (SL-VAE) to jointly derive (layouts, semantics) from the input scene graph, which allows a more diverse and reasonable generation in a one-to-many mapping. We then develop a Compositional Masked Attention (CMA) integrated with a diffusion model, incorporating (layouts, semantics) with fine-grained attributes as generation guidance. To further achieve graph manipulation while keeping the visual content consistent, we introduce a Multi-Layered Sampler (MLS) for an "isolated" image editing effect. Extensive experiments demonstrate that our method outperforms recent competitors based on text, layout, or scene graph, in terms of generation rationality and controllability.

Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models

Disentangled representation learning (DRL) aims to identify and decompose underlying factors behind observations, thus facilitating data perception and generation. However, current DRL approaches often rely on the unrealistic assumption that semantic factors are statistically independent. In reality, these factors may exhibit correlations, which off-the-shelf solutions have yet to properly address. To tackle this challenge, we introduce a bidirectional weighted graph-based framework, to learn factorized attributes and their interrelations within complex data. Specifically, we propose a $\beta$-VAE based module to extract factors as the initial nodes of the graph, and leverage the multimodal large language model (MLLM) to discover and rank latent correlations, thereby updating the weighted edges. By integrating these complementary modules, our model successfully achieves fine-grained, practical and unsupervised disentanglement. Experiments demonstrate our method's superior performance in disentanglement and reconstruction. Furthermore, the model inherits enhanced interpretability and generalizability from MLLMs.

One at A Time: Multi-step Volumetric Probability Distribution Diffusion for Depth Estimation

Recent works have explored the fundamental role of depth estimation in multi-view stereo (MVS) and semantic scene completion (SSC). They generally construct 3D cost volumes to explore geometric correspondence in depth, and estimate such volumes in a single step relying directly on the ground truth approximation. However, such problem cannot be thoroughly handled in one step due to complex empirical distributions, especially in challenging regions like occlusions, reflections, etc. In this paper, we formulate the depth estimation task as a multi-step distribution approximation process, and introduce a new paradigm of modeling the Volumetric Probability Distribution progressively (step-by-step) following a Markov chain with Diffusion models (VPDD). Specifically, to constrain the multi-step generation of volume in VPDD, we construct a meta volume guidance and a confidence-aware contextual guidance as conditional geometry priors to facilitate the distribution approximation. For the sampling process, we further investigate an online filtering strategy to maintain consistency in volume representations for stable training. Experiments demonstrate that our plug-and-play VPDD outperforms the state-of-the-arts for tasks of MVS and SSC, and can also be easily extended to different baselines to get improvement. It is worth mentioning that we are the first camera-based work that surpasses LiDAR-based methods on the SemanticKITTI dataset.