Prior point cloud provides 3D environmental context, which enhances the capabilities of monocular camera in downstream vision tasks, such as 3D object detection, via data fusion. However, the absence of accurate and automated registration methods for estimating camera extrinsic parameters in roadside scene point clouds notably constrains the potential applications of roadside cameras. This paper proposes a novel approach for the automatic registration between prior point clouds and images from roadside scenes. The main idea involves rendering photorealistic grayscale views taken at specific perspectives from the prior point cloud with the help of their features like RGB or intensity values. These generated views can reduce the modality differences between images and prior point clouds, thereby improve the robustness and accuracy of the registration results. Particularly, we specify an efficient algorithm, named neighbor rendering, for the rendering process. Then we introduce a method for automatically estimating the initial guess using only rough guesses of camera's position. At last, we propose a procedure for iteratively refining the extrinsic parameters by minimizing the reprojection error for line features extracted from both generated and camera images using Segment Anything Model (SAM). We assess our method using a self-collected dataset, comprising eight cameras strategically positioned throughout the university campus. Experiments demonstrate our method's capability to automatically align prior point cloud with roadside camera image, achieving a rotation accuracy of 0.202 degrees and a translation precision of 0.079m. Furthermore, we validate our approach's effectiveness in visual applications by substantially improving monocular 3D object detection performance.
Localization and mapping are critical tasks for various applications such as autonomous vehicles and robotics. The challenges posed by outdoor environments present particular complexities due to their unbounded characteristics. In this work, we present MM-Gaussian, a LiDAR-camera multi-modal fusion system for localization and mapping in unbounded scenes. Our approach is inspired by the recently developed 3D Gaussians, which demonstrate remarkable capabilities in achieving high rendering quality and fast rendering speed. Specifically, our system fully utilizes the geometric structure information provided by solid-state LiDAR to address the problem of inaccurate depth encountered when relying solely on visual solutions in unbounded, outdoor scenarios. Additionally, we utilize 3D Gaussian point clouds, with the assistance of pixel-level gradient descent, to fully exploit the color information in photos, thereby achieving realistic rendering effects. To further bolster the robustness of our system, we designed a relocalization module, which assists in returning to the correct trajectory in the event of a localization failure. Experiments conducted in multiple scenarios demonstrate the effectiveness of our method.
Numerous roadside perception datasets have been introduced to propel advancements in autonomous driving and intelligent transportation systems research and development. However, it has been observed that the majority of their concentrates is on urban arterial roads, inadvertently overlooking residential areas such as parks and campuses that exhibit entirely distinct characteristics. In light of this gap, we propose CORP, which stands as the first public benchmark dataset tailored for multi-modal roadside perception tasks under campus scenarios. Collected in a university campus, CORP consists of over 205k images plus 102k point clouds captured from 18 cameras and 9 LiDAR sensors. These sensors with different configurations are mounted on roadside utility poles to provide diverse viewpoints within the campus region. The annotations of CORP encompass multi-dimensional information beyond 2D and 3D bounding boxes, providing extra support for 3D seamless tracking and instance segmentation with unique IDs and pixel masks for identifying targets, to enhance the understanding of objects and their behaviors distributed across the campus premises. Unlike other roadside datasets about urban traffic, CORP extends the spectrum to highlight the challenges for multi-modal perception in campuses and other residential areas.
The integration of complementary characteristics from camera and radar data has emerged as an effective approach in 3D object detection. However, such fusion-based methods remain unexplored for place recognition, an equally important task for autonomous systems. Given that place recognition relies on the similarity between a query scene and the corresponding candidate scene, the stationary background of a scene is expected to play a crucial role in the task. As such, current well-designed camera-radar fusion methods for 3D object detection can hardly take effect in place recognition because they mainly focus on dynamic foreground objects. In this paper, a background-attentive camera-radar fusion-based method, named CRPlace, is proposed to generate background-attentive global descriptors from multi-view images and radar point clouds for accurate place recognition. To extract stationary background features effectively, we design an adaptive module that generates the background-attentive mask by utilizing the camera BEV feature and radar dynamic points. With the guidance of a background mask, we devise a bidirectional cross-attention-based spatial fusion strategy to facilitate comprehensive spatial interaction between the background information of the camera BEV feature and the radar BEV feature. As the first camera-radar fusion-based place recognition network, CRPlace has been evaluated thoroughly on the nuScenes dataset. The results show that our algorithm outperforms a variety of baseline methods across a comprehensive set of metrics (recall@1 reaches 91.2%).
The ability to understand and reason the 3D real world is a crucial milestone towards artificial general intelligence. The current common practice is to finetune Large Language Models (LLMs) with 3D data and texts to enable 3D understanding. Despite their effectiveness, these approaches are inherently limited by the scale and diversity of the available 3D data. Alternatively, in this work, we introduce Agent3D-Zero, an innovative 3D-aware agent framework addressing the 3D scene understanding in a zero-shot manner. The essence of our approach centers on reconceptualizing the challenge of 3D scene perception as a process of understanding and synthesizing insights from multiple images, inspired by how our human beings attempt to understand 3D scenes. By consolidating this idea, we propose a novel way to make use of a Large Visual Language Model (VLM) via actively selecting and analyzing a series of viewpoints for 3D understanding. Specifically, given an input 3D scene, Agent3D-Zero first processes a bird's-eye view image with custom-designed visual prompts, then iteratively chooses the next viewpoints to observe and summarize the underlying knowledge. A distinctive advantage of Agent3D-Zero is the introduction of novel visual prompts, which significantly unleash the VLMs' ability to identify the most informative viewpoints and thus facilitate observing 3D scenes. Extensive experiments demonstrate the effectiveness of the proposed framework in understanding diverse and previously unseen 3D environments.
We present a hybrid-view-based knowledge distillation framework, termed HVDistill, to guide the feature learning of a point cloud neural network with a pre-trained image network in an unsupervised manner. By exploiting the geometric relationship between RGB cameras and LiDAR sensors, the correspondence between the two modalities based on both image-plane view and bird-eye view can be established, which facilitates representation learning. Specifically, the image-plane correspondences can be simply obtained by projecting the point clouds, while the bird-eye-view correspondences can be achieved by lifting pixels to the 3D space with the predicted depths under the supervision of projected point clouds. The image teacher networks provide rich semantics from the image-plane view and meanwhile acquire geometric information from the bird-eye view. Indeed, image features from the two views naturally complement each other and together can ameliorate the learned feature representation of the point cloud student networks. Moreover, with a self-supervised pre-trained 2D network, HVDistill requires neither 2D nor 3D annotations. We pre-train our model on nuScenes dataset and transfer it to several downstream tasks on nuScenes, SemanticKITTI, and KITTI datasets for evaluation. Extensive experimental results show that our method achieves consistent improvements over the baseline trained from scratch and significantly outperforms the existing schemes. Codes are available at git@github.com:zhangsha1024/HVDistill.git.
In this work, we present PoIFusion, a simple yet effective multi-modal 3D object detection framework to fuse the information of RGB images and LiDAR point clouds at the point of interest (abbreviated as PoI). Technically, our PoIFusion follows the paradigm of query-based object detection, formulating object queries as dynamic 3D boxes. The PoIs are adaptively generated from each query box on the fly, serving as the keypoints to represent a 3D object and play the role of basic units in multi-modal fusion. Specifically, we project PoIs into the view of each modality to sample the corresponding feature and integrate the multi-modal features at each PoI through a dynamic fusion block. Furthermore, the features of PoIs derived from the same query box are aggregated together to update the query feature. Our approach prevents information loss caused by view transformation and eliminates the computation-intensive global attention, making the multi-modal 3D object detector more applicable. We conducted extensive experiments on the nuScenes dataset to evaluate our approach. Remarkably, our PoIFusion achieves 74.9\% NDS and 73.4\% mAP, setting a state-of-the-art record on the multi-modal 3D object detection benchmark. Codes will be made available via \url{https://djiajunustc.github.io/projects/poifusion}.
Place recognition is crucial for tasks like loop-closure detection and re-localization. Single-chip millimeter wave radar (single-chip radar in short) emerges as a low-cost sensor option for place recognition, with the advantage of insensitivity to degraded visual environments. However, it encounters two challenges. Firstly, sparse point cloud from single-chip radar leads to poor performance when using current place recognition methods, which assume much denser data. Secondly, its performance significantly declines in scenarios involving rotational and lateral variations, due to limited overlap in its field of view (FOV). We propose mmPlace, a robust place recognition system to address these challenges. Specifically, mmPlace transforms intermediate frequency (IF) signal into range azimuth heatmap and employs a spatial encoder to extract features. Additionally, to improve the performance in scenarios involving rotational and lateral variations, mmPlace employs a rotating platform and concatenates heatmaps in a rotation cycle, effectively expanding the system's FOV. We evaluate mmPlace's performance on the milliSonic dataset, which is collected on the University of Science and Technology of China (USTC) campus, the city roads surrounding the campus, and an underground parking garage. The results demonstrate that mmPlace outperforms point cloud-based methods and achieves 87.37% recall@1 in scenarios involving rotational and lateral variations.
Graph Convolutional Networks (GCNs) have become pivotal in recommendation systems for learning user and item embeddings by leveraging the user-item interaction graph's node information and topology. However, these models often face the famous over-smoothing issue, leading to indistinct user and item embeddings and reduced personalization. Traditional desmoothing methods in GCN-based systems are model-specific, lacking a universal solution. This paper introduces a novel, model-agnostic approach named \textbf{D}esmoothing Framework for \textbf{G}CN-based \textbf{R}ecommendation Systems (\textbf{DGR}). It effectively addresses over-smoothing on general GCN-based recommendation models by considering both global and local perspectives. Specifically, we first introduce vector perturbations during each message passing layer to penalize the tendency of node embeddings approximating overly to be similar with the guidance of the global topological structure. Meanwhile, we further develop a tailored-design loss term for the readout embeddings to preserve the local collaborative relations between users and their neighboring items. In particular, items that exhibit a high correlation with neighboring items are also incorporated to enhance the local topological information. To validate our approach, we conduct extensive experiments on 5 benchmark datasets based on 5 well-known GCN-based recommendation models, demonstrating the effectiveness and generalization of our proposed framework.
The fusion of LiDARs and cameras has been increasingly adopted in autonomous driving for perception tasks. The performance of such fusion-based algorithms largely depends on the accuracy of sensor calibration, which is challenging due to the difficulty of identifying common features across different data modalities. Previously, many calibration methods involved specific targets and/or manual intervention, which has proven to be cumbersome and costly. Learning-based online calibration methods have been proposed, but their performance is barely satisfactory in most cases. These methods usually suffer from issues such as sparse feature maps, unreliable cross-modality association, inaccurate calibration parameter regression, etc. In this paper, to address these issues, we propose CalibFormer, an end-to-end network for automatic LiDAR-camera calibration. We aggregate multiple layers of camera and LiDAR image features to achieve high-resolution representations. A multi-head correlation module is utilized to identify correlations between features more accurately. Lastly, we employ transformer architectures to estimate accurate calibration parameters from the correlation information. Our method achieved a mean translation error of $0.8751 \mathrm{cm}$ and a mean rotation error of $0.0562 ^{\circ}$ on the KITTI dataset, surpassing existing state-of-the-art methods and demonstrating strong robustness, accuracy, and generalization capabilities.