Unleashing the Potential of Multimodal LLMs for Zero-Shot Spatio-Temporal Video Grounding

Spatio-temporal video grounding (STVG) aims at localizing the spatio-temporal tube of a video, as specified by the input text query. In this paper, we utilize multimodal large language models (MLLMs) to explore a zero-shot solution in STVG. We reveal two key insights about MLLMs: (1) MLLMs tend to dynamically assign special tokens, referred to as \textit{grounding tokens}, for grounding the text query; and (2) MLLMs often suffer from suboptimal grounding due to the inability to fully integrate the cues in the text query (\textit{e.g.}, attributes, actions) for inference. Based on these insights, we propose a MLLM-based zero-shot framework for STVG, which includes novel decomposed spatio-temporal highlighting (DSTH) and temporal-augmented assembling (TAS) strategies to unleash the reasoning ability of MLLMs. The DSTH strategy first decouples the original query into attribute and action sub-queries for inquiring the existence of the target both spatially and temporally. It then uses a novel logit-guided re-attention (LRA) module to learn latent variables as spatial and temporal prompts, by regularizing token predictions for each sub-query. These prompts highlight attribute and action cues, respectively, directing the model's attention to reliable spatial and temporal related visual regions. In addition, as the spatial grounding by the attribute sub-query should be temporally consistent, we introduce the TAS strategy to assemble the predictions using the original video frames and the temporal-augmented frames as inputs to help improve temporal consistency. We evaluate our method on various MLLMs, and show that it outperforms SOTA methods on three common STVG benchmarks. The code will be available at https://github.com/zaiquanyang/LLaVA_Next_STVG.

StyleSculptor: Zero-Shot Style-Controllable 3D Asset Generation with Texture-Geometry Dual Guidance

Creating 3D assets that follow the texture and geometry style of existing ones is often desirable or even inevitable in practical applications like video gaming and virtual reality. While impressive progress has been made in generating 3D objects from text or images, creating style-controllable 3D assets remains a complex and challenging problem. In this work, we propose StyleSculptor, a novel training-free approach for generating style-guided 3D assets from a content image and one or more style images. Unlike previous works, StyleSculptor achieves style-guided 3D generation in a zero-shot manner, enabling fine-grained 3D style control that captures the texture, geometry, or both styles of user-provided style images. At the core of StyleSculptor is a novel Style Disentangled Attention (SD-Attn) module, which establishes a dynamic interaction between the input content image and style image for style-guided 3D asset generation via a cross-3D attention mechanism, enabling stable feature fusion and effective style-guided generation. To alleviate semantic content leakage, we also introduce a style-disentangled feature selection strategy within the SD-Attn module, which leverages the variance of 3D feature patches to disentangle style- and content-significant channels, allowing selective feature injection within the attention framework. With SD-Attn, the network can dynamically compute texture-, geometry-, or both-guided features to steer the 3D generation process. Built upon this, we further propose the Style Guided Control (SGC) mechanism, which enables exclusive geometry- or texture-only stylization, as well as adjustable style intensity control. Extensive experiments demonstrate that StyleSculptor outperforms existing baseline methods in producing high-fidelity 3D assets.

HOComp: Interaction-Aware Human-Object Composition

While existing image-guided composition methods may help insert a foreground object onto a user-specified region of a background image, achieving natural blending inside the region with the rest of the image unchanged, we observe that these existing methods often struggle in synthesizing seamless interaction-aware compositions when the task involves human-object interactions. In this paper, we first propose HOComp, a novel approach for compositing a foreground object onto a human-centric background image, while ensuring harmonious interactions between the foreground object and the background person and their consistent appearances. Our approach includes two key designs: (1) MLLMs-driven Region-based Pose Guidance (MRPG), which utilizes MLLMs to identify the interaction region as well as the interaction type (e.g., holding and lefting) to provide coarse-to-fine constraints to the generated pose for the interaction while incorporating human pose landmarks to track action variations and enforcing fine-grained pose constraints; and (2) Detail-Consistent Appearance Preservation (DCAP), which unifies a shape-aware attention modulation mechanism, a multi-view appearance loss, and a background consistency loss to ensure consistent shapes/textures of the foreground and faithful reproduction of the background human. We then propose the first dataset, named Interaction-aware Human-Object Composition (IHOC), for the task. Experimental results on our dataset show that HOComp effectively generates harmonious human-object interactions with consistent appearances, and outperforms relevant methods qualitatively and quantitatively.

Voyager: Long-Range and World-Consistent Video Diffusion for Explorable 3D Scene Generation

Real-world applications like video gaming and virtual reality often demand the ability to model 3D scenes that users can explore along custom camera trajectories. While significant progress has been made in generating 3D objects from text or images, creating long-range, 3D-consistent, explorable 3D scenes remains a complex and challenging problem. In this work, we present Voyager, a novel video diffusion framework that generates world-consistent 3D point-cloud sequences from a single image with user-defined camera path. Unlike existing approaches, Voyager achieves end-to-end scene generation and reconstruction with inherent consistency across frames, eliminating the need for 3D reconstruction pipelines (e.g., structure-from-motion or multi-view stereo). Our method integrates three key components: 1) World-Consistent Video Diffusion: A unified architecture that jointly generates aligned RGB and depth video sequences, conditioned on existing world observation to ensure global coherence 2) Long-Range World Exploration: An efficient world cache with point culling and an auto-regressive inference with smooth video sampling for iterative scene extension with context-aware consistency, and 3) Scalable Data Engine: A video reconstruction pipeline that automates camera pose estimation and metric depth prediction for arbitrary videos, enabling large-scale, diverse training data curation without manual 3D annotations. Collectively, these designs result in a clear improvement over existing methods in visual quality and geometric accuracy, with versatile applications.

Hierarchical Cross-Modal Alignment for Open-Vocabulary 3D Object Detection

Open-vocabulary 3D object detection (OV-3DOD) aims at localizing and classifying novel objects beyond closed sets. The recent success of vision-language models (VLMs) has demonstrated their remarkable capabilities to understand open vocabularies. Existing works that leverage VLMs for 3D object detection (3DOD) generally resort to representations that lose the rich scene context required for 3D perception. To address this problem, we propose in this paper a hierarchical framework, named HCMA, to simultaneously learn local object and global scene information for OV-3DOD. Specifically, we first design a Hierarchical Data Integration (HDI) approach to obtain coarse-to-fine 3D-image-text data, which is fed into a VLM to extract object-centric knowledge. To facilitate the association of feature hierarchies, we then propose an Interactive Cross-Modal Alignment (ICMA) strategy to establish effective intra-level and inter-level feature connections. To better align features across different levels, we further propose an Object-Focusing Context Adjustment (OFCA) module to refine multi-level features by emphasizing object-related features. Extensive experiments demonstrate that the proposed method outperforms SOTA methods on the existing OV-3DOD benchmarks. It also achieves promising OV-3DOD results even without any 3D annotations.

Do Multimodal Large Language Models See Like Humans?

Multimodal Large Language Models (MLLMs) have achieved impressive results on various vision tasks, leveraging recent advancements in large language models. However, a critical question remains unaddressed: do MLLMs perceive visual information similarly to humans? Current benchmarks lack the ability to evaluate MLLMs from this perspective. To address this challenge, we introduce HVSBench, a large-scale benchmark designed to assess the alignment between MLLMs and the human visual system (HVS) on fundamental vision tasks that mirror human vision. HVSBench curated over 85K multimodal samples, spanning 13 categories and 5 fields in HVS, including Prominence, Subitizing, Prioritizing, Free-Viewing, and Searching. Extensive experiments demonstrate the effectiveness of our benchmark in providing a comprehensive evaluation of MLLMs. Specifically, we evaluate 13 MLLMs, revealing that even the best models show significant room for improvement, with most achieving only moderate results. Our experiments reveal that HVSBench presents a new and significant challenge for cutting-edge MLLMs. We believe that HVSBench will facilitate research on human-aligned and explainable MLLMs, marking a key step in understanding how MLLMs perceive and process visual information.

Revisiting the Integration of Convolution and Attention for Vision Backbone

Convolutions (Convs) and multi-head self-attentions (MHSAs) are typically considered alternatives to each other for building vision backbones. Although some works try to integrate both, they apply the two operators simultaneously at the finest pixel granularity. With Convs responsible for per-pixel feature extraction already, the question is whether we still need to include the heavy MHSAs at such a fine-grained level. In fact, this is the root cause of the scalability issue w.r.t. the input resolution for vision transformers. To address this important problem, we propose in this work to use MSHAs and Convs in parallel \textbf{at different granularity levels} instead. Specifically, in each layer, we use two different ways to represent an image: a fine-grained regular grid and a coarse-grained set of semantic slots. We apply different operations to these two representations: Convs to the grid for local features, and MHSAs to the slots for global features. A pair of fully differentiable soft clustering and dispatching modules is introduced to bridge the grid and set representations, thus enabling local-global fusion. Through extensive experiments on various vision tasks, we empirically verify the potential of the proposed integration scheme, named \textit{GLMix}: by offloading the burden of fine-grained features to light-weight Convs, it is sufficient to use MHSAs in a few (e.g., 64) semantic slots to match the performance of recent state-of-the-art backbones, while being more efficient. Our visualization results also demonstrate that the soft clustering module produces a meaningful semantic grouping effect with only IN1k classification supervision, which may induce better interpretability and inspire new weakly-supervised semantic segmentation approaches. Code will be available at \url{https://github.com/rayleizhu/GLMix}.

LuSh-NeRF: Lighting up and Sharpening NeRFs for Low-light Scenes

Neural Radiance Fields (NeRFs) have shown remarkable performances in producing novel-view images from high-quality scene images. However, hand-held low-light photography challenges NeRFs as the captured images may simultaneously suffer from low visibility, noise, and camera shakes. While existing NeRF methods may handle either low light or motion, directly combining them or incorporating additional image-based enhancement methods does not work as these degradation factors are highly coupled. We observe that noise in low-light images is always sharp regardless of camera shakes, which implies an implicit order of these degradation factors within the image formation process. To this end, we propose in this paper a novel model, named LuSh-NeRF, which can reconstruct a clean and sharp NeRF from a group of hand-held low-light images. The key idea of LuSh-NeRF is to sequentially model noise and blur in the images via multi-view feature consistency and frequency information of NeRF, respectively. Specifically, LuSh-NeRF includes a novel Scene-Noise Decomposition (SND) module for decoupling the noise from the scene representation and a novel Camera Trajectory Prediction (CTP) module for the estimation of camera motions based on low-frequency scene information. To facilitate training and evaluations, we construct a new dataset containing both synthetic and real images. Experiments show that LuSh-NeRF outperforms existing approaches. Our code and dataset can be found here: https://github.com/quzefan/LuSh-NeRF.

Boosting Weakly-Supervised Referring Image Segmentation via Progressive Comprehension

This paper explores the weakly-supervised referring image segmentation (WRIS) problem, and focuses on a challenging setup where target localization is learned directly from image-text pairs. We note that the input text description typically already contains detailed information on how to localize the target object, and we also observe that humans often follow a step-by-step comprehension process (\ie, progressively utilizing target-related attributes and relations as cues) to identify the target object. Hence, we propose a novel Progressive Comprehension Network (PCNet) to leverage target-related textual cues from the input description for progressively localizing the target object. Specifically, we first use a Large Language Model (LLM) to decompose the input text description into short phrases. These short phrases are taken as target-related cues and fed into a Conditional Referring Module (CRM) in multiple stages, to allow updating the referring text embedding and enhance the response map for target localization in a multi-stage manner. Based on the CRM, we then propose a Region-aware Shrinking (RaS) loss to constrain the visual localization to be conducted progressively in a coarse-to-fine manner across different stages. Finally, we introduce an Instance-aware Disambiguation (IaD) loss to suppress instance localization ambiguity by differentiating overlapping response maps generated by different referring texts on the same image. Extensive experiments show that our method outperforms SOTA methods on three common benchmarks.

Phidias: A Generative Model for Creating 3D Content from Text, Image, and 3D Conditions with Reference-Augmented Diffusion

In 3D modeling, designers often use an existing 3D model as a reference to create new ones. This practice has inspired the development of Phidias, a novel generative model that uses diffusion for reference-augmented 3D generation. Given an image, our method leverages a retrieved or user-provided 3D reference model to guide the generation process, thereby enhancing the generation quality, generalization ability, and controllability. Our model integrates three key components: 1) meta-ControlNet that dynamically modulates the conditioning strength, 2) dynamic reference routing that mitigates misalignment between the input image and 3D reference, and 3) self-reference augmentations that enable self-supervised training with a progressive curriculum. Collectively, these designs result in a clear improvement over existing methods. Phidias establishes a unified framework for 3D generation using text, image, and 3D conditions with versatile applications.