Audio synthesis has broad applications in multimedia. Recent advancements have made it possible to generate relevant audios from inputs describing an audio scene, such as images or texts. However, the immersiveness and expressiveness of the generation are limited. One possible problem is that existing methods solely rely on the global scene and overlook details of local sounding objects (i.e., sound sources). To address this issue, we propose a Sound Source-Aware Audio (SS2A) generator. SS2A is able to locally perceive multimodal sound sources from a scene with visual detection and cross-modality translation. It then contrastively learns a Cross-Modal Sound Source (CMSS) Manifold to semantically disambiguate each source. Finally, we attentively mix their CMSS semantics into a rich audio representation, from which a pretrained audio generator outputs the sound. To model the CMSS manifold, we curate a novel single-sound-source visual-audio dataset VGGS3 from VGGSound. We also design a Sound Source Matching Score to clearly measure localized audio relevance. With the effectiveness of explicit sound source modeling, SS2A achieves state-of-the-art performance in extensive image-to-audio tasks. We also qualitatively demonstrate SS2A's ability to achieve intuitive synthesis control by compositing vision, text, and audio conditions. Furthermore, we show that our sound source modeling can achieve competitive video-to-audio performance with a straightforward temporal aggregation mechanism.

Modern artistic productions increasingly demand automated choreography generation that adapts to diverse musical styles and individual dancer characteristics. Existing approaches often fail to produce high-quality dance videos that harmonize with both musical rhythm and user-defined choreography styles, limiting their applicability in real-world creative contexts. To address this gap, we introduce ChoreoMuse, a diffusion-based framework that uses SMPL format parameters and their variation version as intermediaries between music and video generation, thereby overcoming the usual constraints imposed by video resolution. Critically, ChoreoMuse supports style-controllable, high-fidelity dance video generation across diverse musical genres and individual dancer characteristics, including the flexibility to handle any reference individual at any resolution. Our method employs a novel music encoder MotionTune to capture motion cues from audio, ensuring that the generated choreography closely follows the beat and expressive qualities of the input music. To quantitatively evaluate how well the generated dances match both musical and choreographic styles, we introduce two new metrics that measure alignment with the intended stylistic cues. Extensive experiments confirm that ChoreoMuse achieves state-of-the-art performance across multiple dimensions, including video quality, beat alignment, dance diversity, and style adherence, demonstrating its potential as a robust solution for a wide range of creative applications. Video results can be found on our project page: https://choreomuse.github.io.

The primary challenge in deploying Human Pose Estimation (HPE) methods in real-world applications lies in balancing computational speed, model compactness, and prediction accuracy. Existing methods achieve strong performance in one or two aspects, but usually at the expense of the remaining one. To overcome this trade-off, we propose KeyRegionPose, a novel framework that achieves high accuracy while reducing model size and computational cost. Central to our design is the Region Focus Mechanism, which enables the model to concentrate on keypoint-relevant regions rather than the entire image. During training, we generate intermediate keypoint proposals to estimate keypoint-specific areas, from which the model learns region-focused features and refines predictions. To ensure accurate keypoint localization and enhance final pose estimation performance, we introduce a Cross-Representation Consistency Loss (CRC Loss) that enforces alignment between the predicted heatmaps and the regressed keypoint coordinates. Additionally, we propose Progressive Multi-Scale Token Pruning (PMTP), a strategy that prunes irrelevant tokens across multiple feature scales to accelerate inference. KeyRegionPose achieves 76.0 AP on the COCO validation set and 75.4 AP on the test-dev set, with only 20.0 million parameters and 8.6 GFLOPs—representing a 27.3% reduction in parameter count, 21.8% decrease in GFLOPs, and a competitive result (+0.2%) over state-of-the-art lightweight HPE models.

Generating dance from music is crucial for advancing automated choreography. Current methods typically produce skeleton keypoint sequences instead of dance videos and lack the capability to make specific individuals dance, which reduces their real-world applicability. These methods also require precise keypoint annotations, complicating data collection and limiting the use of self-collected video datasets. To overcome these challenges, we introduce a novel task: generating dance videos directly from images of individuals guided by music. This task enables the dance generation of specific individuals without requiring keypoint annotations, making it more versatile and applicable to various situations. Our solution, the Dance Any Beat Diffusion model (DabFusion), utilizes a reference image and a music piece to generate dance videos featuring various dance types and choreographies. The music is analyzed by our specially designed music encoder, which identifies essential features including dance style, movement, and rhythm. DabFusion excels in generating dance videos not only for individuals in the training dataset but also for any previously unseen person. This versatility stems from its approach of generating latent optical flow, which contains all necessary motion information to animate any person in the image. We evaluate DabFusion's performance using the AIST + + dataset, focusing on video quality, audio-video synchronization, and motion-music alignment. We propose a 2D Motion-Music Alignment Score (2D-MM Align), which builds on the Beat Alignment Score to more effectively evaluate motion-music alignment for this new task. Experiments show that our DabFusion establishes a solid baseline for this innovative task. Video results can be found on our project page: https://DabFusion.github.io.

Building artificial intelligence (AI) systems on top of a set of foundation models (FMs) is becoming a new paradigm in AI research. Their representative and generative abilities learnt from vast amounts of data can be easily adapted and transferred to a wide range of downstream tasks without extra training from scratch. However, leveraging FMs in cross-modal generation remains under-researched when audio modality is involved. On the other hand, automatically generating semantically-relevant sound from visual input is an important problem in cross-modal generation studies. To solve this vision-to-audio (V2A) generation problem, existing methods tend to design and build complex systems from scratch using modestly sized datasets. In this paper, we propose a lightweight solution to this problem by leveraging foundation models, specifically CLIP, CLAP, and AudioLDM. We first investigate the domain gap between the latent space of the visual CLIP and the auditory CLAP models. Then we propose a simple yet effective mapper mechanism (V2A-Mapper) to bridge the domain gap by translating the visual input between CLIP and CLAP spaces. Conditioned on the translated CLAP embedding, pretrained audio generative FM AudioLDM is adopted to produce high-fidelity and visually-aligned sound. Compared to previous approaches, our method only requires a quick training of the V2A-Mapper. We further analyze and conduct extensive experiments on the choice of the V2A-Mapper and show that a generative mapper is better at fidelity and variability (FD) while a regression mapper is slightly better at relevance (CS). Both objective and subjective evaluation on two V2A datasets demonstrate the superiority of our proposed method compared to current state-of-the-art approaches - trained with 86% fewer parameters but achieving 53% and 19% improvement in FD and CS, respectively. Supplementary materials such as audio samples are provided at our demo website: https://v2a-mapper.github.io/.

Recent advancements in Vision-Language (VL) research have sparked new benchmarks for complex visual reasoning, challenging models’ advanced reasoning ability. Traditional Vision-Language models (VLMs) perform well in visual perception tasks while struggling with complex reasoning scenarios. Conversely, Large Language Models (LLMs) demonstrate robust text reasoning capabilities; however, they lack visual acuity. To bridge this gap, we propose Complex Visual Reasoning Large Language Models (CVR-LLM), capitalizing on VLMs’ visual perception proficiency and LLMs’ extensive reasoning capability. Unlike recent multimodal large language models (MLLMs) that require a projection layer, our approach transforms images into detailed, context-aware descriptions using an iterative self-refinement loop and leverages LLMs’ text knowledge for accurate predictions without extra training. We also introduce a novel multi-modal in-context learning (ICL) methodology to enhance LLMs’ contextual understanding and reasoning. Additionally, we introduce Chain-of-Comparison (CoC), a step-by-step comparison technique enabling contrasting various aspects of predictions. Our CVR-LLM presents the first comprehensive study across a wide array of complex visual reasoning tasks and achieves SOTA performance among all.

Advanced Driver-Assistance Systems (ADAS) have successfully integrated learning-based techniques into vehicle perception and decision-making. However, their application in 3D lane detection for effective driving environment perception is hindered by the lack of comprehensive LiDAR datasets. The sparse nature of LiDAR point cloud data prevents an efficient manual annotation process. To solve this problem, we present LiSV-3DLane, a large-scale 3D lane dataset that comprises 20k frames of surround-view LiDAR point clouds with enriched semantic annotation. Unlike existing datasets confined to a frontal perspective, LiSV-3DLane provides a full 360-degree spatial panorama around the ego vehicle, capturing complex lane patterns in both urban and highway environments. We leverage the geometric traits of lane lines and the intrinsic spatial attributes of LiDAR data to design a simple yet effective automatic annotation pipeline for generating finer lane labels. To propel future research, we propose a novel LiDAR-based 3D lane detection model, LiLaDet, incorporating the spatial geometry learning of the LiDAR point cloud into Bird’s Eye View (BEV) based lane identification. Experimental results indicate that LiLaDet outperforms existing camera- and LiDAR-based approaches in the 3D lane detection task on the K-Lane dataset and our LiSV-3DLane. The project code will be available at https://github.com/RunkaiZhao/LiLaDet.

Detecting 3D lane lines from monocular images is garnering increasing attention in the Autonomous Driving (AD) area due to its cost-effective edge. However, current monocular image models capture road scenes lacking 3D spatial awareness, which is error-prone to adverse circumstance changes. In this work, we design a novel cross-modal knowledge transfer scheme, namely LaneCMKT, to address this challenge by transferring 3D geometric cues learned from a pre-trained LiDAR model to the image model. Performing on the unified Bird's-Eye-View (BEV) grid, our monocular image model acts as a student network and benefits from the spatial guidance of the 3D LiDAR teacher model over the intermediate feature space. Since LiDAR points and image pixels are intrinsically two different modalities, to facilitate such heterogeneous feature transfer learning at matching levels, we propose a dual-path knowledge transfer mechanism. We divide the feature space into shallow and deep paths where the image student model is prompted to focus on lane-favored geometric cues from the LiDAR teacher model. We conduct extensive experiments and thorough analysis on the large-scale public benchmark OpenLane. Our model achieves notable improvements over the image baseline by 5.3% and the current BEV-driven SoTA method by 2.7% in the F1 score, without introducing any extra computational overhead. We also observe that the 3D abilities grabbed from the teacher model are critical for dealing with complex spatial lane properties from a 2D perspective.

Digital neuron reconstruction from 3D microscopy images is an essential technique for investigating brain connectomics and neuron morphology. Existing reconstruction frameworks use convolution-based segmentation networks to partition the neuron from noisy backgrounds before applying the tracing algorithm. The tracing results are sensitive to the raw image quality and segmentation accuracy. In this paper, we propose a novel framework for 3D neuron reconstruction. Our key idea is to use the geometric representation power of the point cloud to better explore the intrinsic structural information of neurons. Our proposed framework adopts one graph convolutional network to predict the neural skeleton points and another one to produce the connectivity of these points. We finally generate the target SWC file through the interpretation of the predicted point coordinates, radius, and connections. Evaluated on the Janelia-Fly dataset from the BigNeuron project, we show that our framework achieves competitive neuron reconstruction performance. Our geometry and topology learning of point clouds could further benefit 3D medical image analysis, such as cardiac surface reconstruction. Our code is available at https://github.com/RunkaiZhao/PointNeuron.

Dense captioning in 3D point clouds is an emerging vision-and-language task involving object-level 3D scene understanding. Apart from coarse semantic class prediction and bounding box regression as in traditional 3D object detection, 3D dense captioning aims at producing a further and finer instance-level label of natural language description on visual appearance and spatial relations for each scene object of interest. To detect and describe objects in a scene, following the spirit of neural machine translation, we propose a transformer-based encoder-decoder architecture, namely SpaCap3D, to transform objects into descriptions, where we especially investigate the relative spatiality of objects in 3D scenes and design a spatiality-guided encoder via a token-to-token spatial relation learning objective and an object-centric decoder for precise and spatiality-enhanced object caption generation. Evaluated on two benchmark datasets, ScanRefer and ReferIt3D, our proposed SpaCap3D outperforms the baseline method Scan2Cap by 4.94% and 9.61% in [email protected], respectively. Our project page with source code and supplementary files is available at https://SpaCap3D.github.io/.