Vision-based roadside 3D object detection has attracted rising attention in autonomous driving domain, since it encompasses inherent advantages in reducing blind spots and expanding perception range. While previous work mainly focuses on accurately estimating depth or height for 2D-to-3D mapping, ignoring the position approximation error in the voxel pooling process. Inspired by this insight, we propose a novel voxel pooling strategy to reduce such error, dubbed BEVSpread. Specifically, instead of bringing the image features contained in a frustum point to a single BEV grid, BEVSpread considers each frustum point as a source and spreads the image features to the surrounding BEV grids with adaptive weights. To achieve superior propagation performance, a specific weight function is designed to dynamically control the decay speed of the weights according to distance and depth. Aided by customized CUDA parallel acceleration, BEVSpread achieves comparable inference time as the original voxel pooling. Extensive experiments on two large-scale roadside benchmarks demonstrate that, as a plug-in, BEVSpread can significantly improve the performance of existing frustum-based BEV methods by a large margin of (1.12, 5.26, 3.01) AP in vehicle, pedestrian and cyclist.
The field of 3D object detection from point clouds is rapidly advancing in computer vision, aiming to accurately and efficiently detect and localize objects in three-dimensional space. Current 3D detectors commonly fall short in terms of flexibility and scalability, with ample room for advancements in performance. In this paper, our objective is to address these limitations by introducing two frameworks for 3D object detection with minimal hand-crafted design. Firstly, we propose CT3D, which sequentially performs raw-point-based embedding, a standard Transformer encoder, and a channel-wise decoder for point features within each proposal. Secondly, we present an enhanced network called CT3D++, which incorporates geometric and semantic fusion-based embedding to extract more valuable and comprehensive proposal-aware information. Additionally, CT3D ++ utilizes a point-to-key bidirectional encoder for more efficient feature encoding with reduced computational cost. By replacing the corresponding components of CT3D with these novel modules, CT3D++ achieves state-of-the-art performance on both the KITTI dataset and the large-scale Way\-mo Open Dataset. The source code for our frameworks will be made accessible at https://github.com/hlsheng1/CT3D-plusplus.
In this work, we propose a disruptively frugal LiDAR perception dataflow that generates rather than senses parts of the environment that are either predictable based on the extensive training of the environment or have limited consequence to the overall prediction accuracy. Therefore, the proposed methodology trades off sensing energy with training data for low-power robotics and autonomous navigation to operate frugally with sensors, extending their lifetime on a single battery charge. Our proposed generative pre-training strategy for this purpose, called as radially masked autoencoding (R-MAE), can also be readily implemented in a typical LiDAR system by selectively activating and controlling the laser power for randomly generated angular regions during on-field operations. Our extensive evaluations show that pre-training with R-MAE enables focusing on the radial segments of the data, thereby capturing spatial relationships and distances between objects more effectively than conventional procedures. Therefore, the proposed methodology not only reduces sensing energy but also improves prediction accuracy. For example, our extensive evaluations on Waymo, nuScenes, and KITTI datasets show that the approach achieves over a 5% average precision improvement in detection tasks across datasets and over a 4% accuracy improvement in transferring domains from Waymo and nuScenes to KITTI. In 3D object detection, it enhances small object detection by up to 4.37% in AP at moderate difficulty levels in the KITTI dataset. Even with 90% radial masking, it surpasses baseline models by up to 5.59% in mAP/mAPH across all object classes in the Waymo dataset. Additionally, our method achieves up to 3.17% and 2.31% improvements in mAP and NDS, respectively, on the nuScenes dataset, demonstrating its effectiveness with both single and fused LiDAR-camera modalities. https://github.com/sinatayebati/Radial_MAE.
3D occupancy prediction (Occ) is a rapidly rising challenging perception task in the field of autonomous driving which represents the driving scene as uniformly partitioned 3D voxel grids with semantics. Compared to 3D object detection, grid perception has great advantage of better recognizing irregularly shaped, unknown category, or partially occluded general objects. However, existing 3D occupancy networks (occnets) are both computationally heavy and label-hungry. In terms of model complexity, occnets are commonly composed of heavy Conv3D modules or transformers on the voxel level. In terms of label annotations requirements, occnets are supervised with large-scale expensive dense voxel labels. Model and data inefficiency, caused by excessive network parameters and label annotations requirement, severely hinder the onboard deployment of occnets. This paper proposes an efficient 3d occupancy network (EFFOcc), that targets the minimal network complexity and label requirement while achieving state-of-the-art accuracy. EFFOcc only uses simple 2D operators, and improves Occ accuracy to the state-of-the-art on multiple large-scale benchmarks: Occ3D-nuScenes, Occ3D-Waymo, and OpenOccupancy-nuScenes. On Occ3D-nuScenes benchmark, EFFOcc has only 18.4M parameters, and achieves 50.46 in terms of mean IoU (mIoU), to our knowledge, it is the occnet with minimal parameters compared with related occnets. Moreover, we propose a two-stage active learning strategy to reduce the requirements of labelled data. Active EFFOcc trained with 6\% labelled voxels achieves 47.19 mIoU, which is 95.7% fully supervised performance. The proposed EFFOcc also supports improved vision-only occupancy prediction with the aid of region-decomposed distillation. Code and demo videos will be available at https://github.com/synsin0/EFFOcc.
Remote sensing lightweight foundation models have achieved notable success in online perception within remote sensing. However, their capabilities are restricted to performing online inference solely based on their own observations and models, thus lacking a comprehensive understanding of large-scale remote sensing scenarios. To overcome this limitation, we propose a Remote Sensing Distributed Foundation Model (RS-DFM) based on generalized information mapping and interaction. This model can realize online collaborative perception across multiple platforms and various downstream tasks by mapping observations into a unified space and implementing a task-agnostic information interaction strategy. Specifically, we leverage the ground-based geometric prior of remote sensing oblique observations to transform the feature mapping from absolute depth estimation to relative depth estimation, thereby enhancing the model's ability to extract generalized features across diverse heights and perspectives. Additionally, we present a dual-branch information compression module to decouple high-frequency and low-frequency feature information, achieving feature-level compression while preserving essential task-agnostic details. In support of our research, we create a multi-task simulation dataset named AirCo-MultiTasks for multi-UAV collaborative observation. We also conduct extensive experiments, including 3D object detection, instance segmentation, and trajectory prediction. The numerous results demonstrate that our RS-DFM achieves state-of-the-art performance across various downstream tasks.
Multi-UAV collaborative 3D object detection can perceive and comprehend complex environments by integrating complementary information, with applications encompassing traffic monitoring, delivery services and agricultural management. However, the extremely broad observations in aerial remote sensing and significant perspective differences across multiple UAVs make it challenging to achieve precise and consistent feature mapping from 2D images to 3D space in multi-UAV collaborative 3D object detection paradigm. To address the problem, we propose an unparalleled camera-based multi-UAV collaborative 3D object detection paradigm called UCDNet. Specifically, the depth information from the UAVs to the ground is explicitly utilized as a strong prior to provide a reference for more accurate and generalizable feature mapping. Additionally, we design a homologous points geometric consistency loss as an auxiliary self-supervision, which directly influences the feature mapping module, thereby strengthening the global consistency of multi-view perception. Experiments on AeroCollab3D and CoPerception-UAVs datasets show our method increases 4.7% and 10% mAP respectively compared to the baseline, which demonstrates the superiority of UCDNet.
With the advancement of collaborative perception, the role of aerial-ground collaborative perception, a crucial component, is becoming increasingly important. The demand for collaborative perception across different perspectives to construct more comprehensive perceptual information is growing. However, challenges arise due to the disparities in the field of view between cross-domain agents and their varying sensitivity to information in images. Additionally, when we transform image features into Bird's Eye View (BEV) features for collaboration, we need accurate depth information. To address these issues, we propose a framework specifically designed for aerial-ground collaboration. First, to mitigate the lack of datasets for aerial-ground collaboration, we develop a virtual dataset named V2U-COO for our research. Second, we design a Cross-Domain Cross-Adaptation (CDCA) module to align the target information obtained from different domains, thereby achieving more accurate perception results. Finally, we introduce a Collaborative Depth Optimization (CDO) module to obtain more precise depth estimation results, leading to more accurate perception outcomes. We conduct extensive experiments on both our virtual dataset and a public dataset to validate the effectiveness of our framework. Our experiments on the V2U-COO dataset and the DAIR-V2X dataset demonstrate that our method improves detection accuracy by 6.1% and 2.7%, respectively.
Next-generation automotive radars provide elevation data in addition to range-, azimuth- and Doppler velocity. In this experimental study, we apply a state-of-the-art object detector (PointPillars), previously used for LiDAR 3D data, to such 3+1D radar data (where 1D refers to Doppler). In ablation studies, we first explore the benefits of the additional elevation information, together with that of Doppler, radar cross section and temporal accumulation, in the context of multi-class road user detection. We subsequently compare object detection performance on the radar and LiDAR point clouds, object class-wise and as a function of distance. To facilitate our experimental study, we present the novel View-of-Delft (VoD) automotive dataset. It contains 8693 frames of synchronized and calibrated 64-layer LiDAR-, (stereo) camera-, and 3+1D radar-data acquired in complex, urban traffic. It consists of 123106 3D bounding box annotations of both moving and static objects, including 26587 pedestrian, 10800 cyclist and 26949 car labels. Our results show that object detection on 64-layer LiDAR data still outperforms that on 3+1D radar data, but the addition of elevation information and integration of successive radar scans helps close the gap. The VoD dataset is made freely available for scientific benchmarking at https://intelligent-vehicles.org/datasets/view-of-delft/ .
Camera and millimeter-wave (MMW) radar fusion is essential for accurate and robust autonomous driving systems. With the advancement of radar technology, next-generation high-resolution automotive radar, i.e., 4-D radar, has emerged. In addition to the target range, azimuth, and Doppler velocity measurements of traditional radar, 4-D radar provides elevation measurement to create a denser “point cloud.” In this study, we propose a camera and 4-D radar fusion network called RCFusion, which achieves multimodal feature fusion under a unified bird’s-eye view (BEV) space to accomplish 3-D object detection tasks. In the camera stream, multiscale feature maps are obtained by the image backbone and feature pyramid network (FPN); they are then converted into orthographic feature maps by an orthographic feature transform (OFT). Next, enhanced and fine-grained image BEV features are obtained via a designed shared attention encoder. Meanwhile, in the 4-D radar stream, a newly designed component named radar PillarNet efficiently encodes the radar features to generate radar pseudo-images, which are fed into the point cloud backbone to create radar BEV features. An interactive attention module (IAM) is proposed for the fusion stage, which outputs a valid fusion of the two-modal BEV features. Finally, a generic detection head predicts the object classes and locations. The proposed RCFusion is validated on the TJ4DRadSet and view-of-delft (VoD) datasets. The experimental results and analysis show that the proposed method can effectively fuse camera and 4-D radar features to achieve robust detection performance.
Open-vocabulary 3D Object Detection (OV-3DDet) addresses the detection of objects from an arbitrary list of novel categories in 3D scenes, which remains a very challenging problem. In this work, we propose CoDAv2, a unified framework designed to innovatively tackle both the localization and classification of novel 3D objects, under the condition of limited base categories. For localization, the proposed 3D Novel Object Discovery (3D-NOD) strategy utilizes 3D geometries and 2D open-vocabulary semantic priors to discover pseudo labels for novel objects during training. 3D-NOD is further extended with an Enrichment strategy that significantly enriches the novel object distribution in the training scenes, and then enhances the model's ability to localize more novel objects. The 3D-NOD with Enrichment is termed 3D-NODE. For classification, the Discovery-driven Cross-modal Alignment (DCMA) module aligns features from 3D point clouds and 2D/textual modalities, employing both class-agnostic and class-specific alignments that are iteratively refined to handle the expanding vocabulary of objects. Besides, 2D box guidance boosts the classification accuracy against complex background noises, which is coined as Box-DCMA. Extensive evaluation demonstrates the superiority of CoDAv2. CoDAv2 outperforms the best-performing method by a large margin (AP_Novel of 9.17 vs. 3.61 on SUN-RGBD and 9.12 vs. 3.74 on ScanNetv2). Source code and pre-trained models are available at the GitHub project page.
With the rapid advancement of autonomous driving technology, there is a growing need for enhanced safety and efficiency in the automatic environmental perception of vehicles during their operation. In modern vehicle setups, cameras and mmWave radar (radar), being the most extensively employed sensors, demonstrate complementary characteristics, inherently rendering them conducive to fusion and facilitating the achievement of both robust performance and cost-effectiveness. This paper focuses on a comprehensive survey of radar-vision (RV) fusion based on deep learning methods for 3D object detection in autonomous driving. We offer a comprehensive overview of each RV fusion category, specifically those employing region of interest (ROI) fusion and end-to-end fusion strategies. As the most promising fusion strategy at present, we provide a deeper classification of end-to-end fusion methods, including those 3D bounding box prediction based and BEV based approaches. Moreover, aligning with recent advancements, we delineate the latest information on 4D radar and its cutting-edge applications in autonomous vehicles (AVs). Finally, we present the possible future trends of RV fusion and summarize this paper.
Cross-modal transformers have demonstrated superiority in various vision tasks by effectively integrating different modalities. This paper first critiques prior token exchange methods which replace less informative tokens with inter-modal features, and demonstrate exchange based methods underperform cross-attention mechanisms, while the computational demand of the latter inevitably restricts its use with longer sequences. To surmount the computational challenges, we propose GeminiFusion, a pixel-wise fusion approach that capitalizes on aligned cross-modal representations. GeminiFusion elegantly combines intra-modal and inter-modal attentions, dynamically integrating complementary information across modalities. We employ a layer-adaptive noise to adaptively control their interplay on a per-layer basis, thereby achieving a harmonized fusion process. Notably, GeminiFusion maintains linear complexity with respect to the number of input tokens, ensuring this multimodal framework operates with efficiency comparable to unimodal networks. Comprehensive evaluations across multimodal image-to-image translation, 3D object detection and arbitrary-modal semantic segmentation tasks, including RGB, depth, LiDAR, event data, etc. demonstrate the superior performance of our GeminiFusion against leading-edge techniques. The PyTorch code is available at https://github.com/JiaDingCN/GeminiFusion
Effective tracking of surrounding traffic participants allows for an accurate state estimation as a necessary ingredient for prediction of future behavior and therefore adequate planning of the ego vehicle trajectory. One approach for detecting and tracking surrounding traffic participants is the combination of a learning based object detector with a classical tracking algorithm. Learning based object detectors have been shown to work adequately on lidar and camera data, while learning based object detectors using standard radar data input have proven to be inferior. Recently, with the improvements to radar sensor technology in the form of imaging radars, the object detection performance on radar was greatly improved but is still limited compared to lidar sensors due to the sparsity of the radar point cloud. This presents a unique challenge for the task of multi-object tracking. The tracking algorithm must overcome the limited detection quality while generating consistent tracks. To this end, a comparison between different multi-object tracking methods on imaging radar data is required to investigate its potential for downstream tasks. The work at hand compares multiple approaches and analyzes their limitations when applied to imaging radar data. Furthermore, enhancements to the presented approaches in the form of probabilistic association algorithms are considered for this task.
Monocular 3D object detection (Mono 3Det) aims to identify 3D objects from a single RGB image. However, existing methods often assume training and test data follow the same distribution, which may not hold in real-world test scenarios. To address the out-of-distribution (OOD) problems, we explore a new adaptation paradigm for Mono 3Det, termed Fully Test-time Adaptation. It aims to adapt a well-trained model to unlabeled test data by handling potential data distribution shifts at test time without access to training data and test labels. However, applying this paradigm in Mono 3Det poses significant challenges due to OOD test data causing a remarkable decline in object detection scores. This decline conflicts with the pre-defined score thresholds of existing detection methods, leading to severe object omissions (i.e., rare positive detections and many false negatives). Consequently, the limited positive detection and plenty of noisy predictions cause test-time adaptation to fail in Mono 3Det. To handle this problem, we propose a novel Monocular Test-Time Adaptation (MonoTTA) method, based on two new strategies. 1) Reliability-driven adaptation: we empirically find that high-score objects are still reliable and the optimization of high-score objects can enhance confidence across all detections. Thus, we devise a self-adaptive strategy to identify reliable objects for model adaptation, which discovers potential objects and alleviates omissions. 2) Noise-guard adaptation: since high-score objects may be scarce, we develop a negative regularization term to exploit the numerous low-score objects via negative learning, preventing overfitting to noise and trivial solutions. Experimental results show that MonoTTA brings significant performance gains for Mono 3Det models in OOD test scenarios, approximately 190% gains by average on KITTI and 198% gains on nuScenes.
Deep neural networks (DNNs) have been proven extremely susceptible to adversarial examples, which raises special safety-critical concerns for DNN-based autonomous driving stacks (i.e., 3D object detection). Although there are extensive works on image-level attacks, most are restricted to 2D pixel spaces, and such attacks are not always physically realistic in our 3D world. Here we present Adv3D, the first exploration of modeling adversarial examples as Neural Radiance Fields (NeRFs). Advances in NeRF provide photorealistic appearances and 3D accurate generation, yielding a more realistic and realizable adversarial example. We train our adversarial NeRF by minimizing the surrounding objects' confidence predicted by 3D detectors on the training set. Then we evaluate Adv3D on the unseen validation set and show that it can cause a large performance reduction when rendering NeRF in any sampled pose. To generate physically realizable adversarial examples, we propose primitive-aware sampling and semantic-guided regularization that enable 3D patch attacks with camouflage adversarial texture. Experimental results demonstrate that the trained adversarial NeRF generalizes well to different poses, scenes, and 3D detectors. Finally, we provide a defense method to our attacks that involves adversarial training through data augmentation. Project page: https://len-li.github.io/adv3d-web
Real-time and high-performance 3D object detection plays a critical role in autonomous driving and robotics. Recent pillar-based 3D object detectors have gained significant attention due to their compact representation and low computational overhead, making them suitable for onboard deployment and quantization. However, existing pillar-based detectors still suffer from information loss along height dimension and large numerical distribution difference during pillar feature encoding (PFE), which severely limits their performance and quantization potential. To address above issue, we first unveil the importance of different input information during PFE and identify the height dimension as a key factor in enhancing 3D detection performance. Motivated by this observation, we propose a height-aware pillar feature encoder, called PillarHist. Specifically, PillarHist statistics the discrete distribution of points at different heights within one pillar with the information entropy guidance. This simple yet effective design greatly preserves the information along the height dimension while significantly reducing the computation overhead of the PFE. Meanwhile, PillarHist also constrains the arithmetic distribution of PFE input to a stable range, making it quantization-friendly. Notably, PillarHist operates exclusively within the PFE stage to enhance performance, enabling seamless integration into existing pillar-based methods without introducing complex operations. Extensive experiments show the effectiveness of PillarHist in terms of both efficiency and performance.
Rapid advancements in Autonomous Driving (AD) tasks turned a significant shift toward end-to-end fashion, particularly in the utilization of vision-language models (VLMs) that integrate robust logical reasoning and cognitive abilities to enable comprehensive end-to-end planning. However, these VLM-based approaches tend to integrate 2D vision tokenizers and a large language model (LLM) for ego-car planning, which lack 3D geometric priors as a cornerstone of reliable planning. Naturally, this observation raises a critical concern: Can a 2D-tokenized LLM accurately perceive the 3D environment? Our evaluation of current VLM-based methods across 3D object detection, vectorized map construction, and environmental caption suggests that the answer is, unfortunately, NO. In other words, 2D-tokenized LLM fails to provide reliable autonomous driving. In response, we introduce DETR-style 3D perceptrons as 3D tokenizers, which connect LLM with a one-layer linear projector. This simple yet elegant strategy, termed Atlas, harnesses the inherent priors of the 3D physical world, enabling it to simultaneously process high-resolution multi-view images and employ spatiotemporal modeling. Despite its simplicity, Atlas demonstrates superior performance in both 3D detection and ego planning tasks on nuScenes dataset, proving that 3D-tokenized LLM is the key to reliable autonomous driving. The code and datasets will be released.
In real-life scenarios, humans seek out objects in the 3D world to fulfill their daily needs or intentions. This inspires us to introduce 3D intention grounding, a new task in 3D object detection employing RGB-D, based on human intention, such as "I want something to support my back". Closely related, 3D visual grounding focuses on understanding human reference. To achieve detection based on human intention, it relies on humans to observe the scene, reason out the target that aligns with their intention ("pillow" in this case), and finally provide a reference to the AI system, such as "A pillow on the couch". Instead, 3D intention grounding challenges AI agents to automatically observe, reason and detect the desired target solely based on human intention. To tackle this challenge, we introduce the new Intent3D dataset, consisting of 44,990 intention texts associated with 209 fine-grained classes from 1,042 scenes of the ScanNet dataset. We also establish several baselines based on different language-based 3D object detection models on our benchmark. Finally, we propose IntentNet, our unique approach, designed to tackle this intention-based detection problem. It focuses on three key aspects: intention understanding, reasoning to identify object candidates, and cascaded adaptive learning that leverages the intrinsic priority logic of different losses for multiple objective optimization.
Constructing large-scale labeled datasets for multi-modal perception model training in autonomous driving presents significant challenges. This has motivated the development of self-supervised pretraining strategies. However, existing pretraining methods mainly employ distinct approaches for each modality. In contrast, we focus on LiDAR-Camera 3D perception models and introduce a unified pretraining strategy, NeRF-Supervised Masked Auto Encoder (NS-MAE), which optimizes all modalities through a shared formulation. NS-MAE leverages NeRF's ability to encode both appearance and geometry, enabling efficient masked reconstruction of multi-modal data. Specifically, embeddings are extracted from corrupted LiDAR point clouds and images, conditioned on view directions and locations. Then, these embeddings are rendered into multi-modal feature maps from two crucial viewpoints for 3D driving perception: perspective and bird's-eye views. The original uncorrupted data serve as reconstruction targets for self-supervised learning. Extensive experiments demonstrate the superior transferability of NS-MAE across various 3D perception tasks under different fine-tuning settings. Notably, NS-MAE outperforms prior SOTA pre-training methods that employ separate strategies for each modality in BEV map segmentation under the label-efficient fine-tuning setting. Our code is publicly available at https://github.com/Xiaohao-Xu/Unified-Pretrain-AD/ .
Ensuring robust 3D object detection and localization is crucial for many applications in robotics and autonomous driving. Recent models, however, face difficulties in maintaining high performance when applied to domains with differing sensor setups or geographic locations, often resulting in poor localization accuracy due to domain shift. To overcome this challenge, we introduce a novel diffusion-based box refinement approach. This method employs a domain-agnostic diffusion model, conditioned on the LiDAR points surrounding a coarse bounding box, to simultaneously refine the box's location, size, and orientation. We evaluate this approach under various domain adaptation settings, and our results reveal significant improvements across different datasets, object classes and detectors. Our PyTorch implementation is available at \href{https://github.com/cxy1997/DiffuBox}{https://github.com/cxy1997/DiffuBox}.