Scaling Multi-Camera 3D Object Detection through Weak-to-Strong Eliciting

The emergence of Multi-Camera 3D Object Detection (MC3D-Det), facilitated by bird's-eye view (BEV) representation, signifies a notable progression in 3D object detection. Scaling MC3D-Det training effectively accommodates varied camera parameters and urban landscapes, paving the way for the MC3D-Det foundation model. However, the multi-view fusion stage of the MC3D-Det method relies on the ill-posed monocular perception during training rather than surround refinement ability, leading to what we term "surround refinement degradation". To this end, our study presents a weak-to-strong eliciting framework aimed at enhancing surround refinement while maintaining robust monocular perception. Specifically, our framework employs weakly tuned experts trained on distinct subsets, and each is inherently biased toward specific camera configurations and scenarios. These biased experts can learn the perception of monocular degeneration, which can help the multi-view fusion stage to enhance surround refinement abilities. Moreover, a composite distillation strategy is proposed to integrate the universal knowledge of 2D foundation models and task-specific information. Finally, for MC3D-Det joint training, the elaborate dataset merge strategy is designed to solve the problem of inconsistent camera numbers and camera parameters. We set up a multiple dataset joint training benchmark for MC3D-Det and adequately evaluated existing methods. Further, we demonstrate the proposed framework brings a generalized and significant boost over multiple baselines. Our code is at \url{https://github.com/EnVision-Research/Scale-BEV}.

GraphAD: Interaction Scene Graph for End-to-end Autonomous Driving

Modeling complicated interactions among the ego-vehicle, road agents, and map elements has been a crucial part for safety-critical autonomous driving. Previous works on end-to-end autonomous driving rely on the attention mechanism for handling heterogeneous interactions, which fails to capture the geometric priors and is also computationally intensive. In this paper, we propose the Interaction Scene Graph (ISG) as a unified method to model the interactions among the ego-vehicle, road agents, and map elements. With the representation of the ISG, the driving agents aggregate essential information from the most influential elements, including the road agents with potential collisions and the map elements to follow. Since a mass of unnecessary interactions are omitted, the more efficient scene-graph-based framework is able to focus on indispensable connections and leads to better performance. We evaluate the proposed method for end-to-end autonomous driving on the nuScenes dataset. Compared with strong baselines, our method significantly outperforms in the full-stack driving tasks, including perception, prediction, and planning. Code will be released at https://github.com/zhangyp15/GraphAD.

Towards Generalizable Multi-Camera 3D Object Detection via Perspective Debiasing

Detecting objects in 3D space using multiple cameras, known as Multi-Camera 3D Object Detection (MC3D-Det), has gained prominence with the advent of bird's-eye view (BEV) approaches. However, these methods often struggle when faced with unfamiliar testing environments due to the lack of diverse training data encompassing various viewpoints and environments. To address this, we propose a novel method that aligns 3D detection with 2D camera plane results, ensuring consistent and accurate detections. Our framework, anchored in perspective debiasing, helps the learning of features resilient to domain shifts. In our approach, we render diverse view maps from BEV features and rectify the perspective bias of these maps, leveraging implicit foreground volumes to bridge the camera and BEV planes. This two-step process promotes the learning of perspective- and context-independent features, crucial for accurate object detection across varying viewpoints, camera parameters and environment conditions. Notably, our model-agnostic approach preserves the original network structure without incurring additional inference costs, facilitating seamless integration across various models and simplifying deployment. Furthermore, we also show our approach achieves satisfactory results in real data when trained only with virtual datasets, eliminating the need for real scene annotations. Experimental results on both Domain Generalization (DG) and Unsupervised Domain Adaptation (UDA) clearly demonstrate its effectiveness. Our code will be released.

OccFormer: Dual-path Transformer for Vision-based 3D Semantic Occupancy Prediction

The vision-based perception for autonomous driving has undergone a transformation from the bird-eye-view (BEV) representations to the 3D semantic occupancy. Compared with the BEV planes, the 3D semantic occupancy further provides structural information along the vertical direction. This paper presents OccFormer, a dual-path transformer network to effectively process the 3D volume for semantic occupancy prediction. OccFormer achieves a long-range, dynamic, and efficient encoding of the camera-generated 3D voxel features. It is obtained by decomposing the heavy 3D processing into the local and global transformer pathways along the horizontal plane. For the occupancy decoder, we adapt the vanilla Mask2Former for 3D semantic occupancy by proposing preserve-pooling and class-guided sampling, which notably mitigate the sparsity and class imbalance. Experimental results demonstrate that OccFormer significantly outperforms existing methods for semantic scene completion on SemanticKITTI dataset and for LiDAR semantic segmentation on nuScenes dataset. Code is available at \url{https://github.com/zhangyp15/OccFormer}.

StereoScene: BEV-Assisted Stereo Matching Empowers 3D Semantic Scene Completion

3D semantic scene completion (SSC) is an ill-posed task that requires inferring a dense 3D scene from incomplete observations. Previous methods either explicitly incorporate 3D geometric input or rely on learnt 3D prior behind monocular RGB images. However, 3D sensors such as LiDAR are expensive and intrusive while monocular cameras face challenges in modeling precise geometry due to the inherent ambiguity. In this work, we propose StereoScene for 3D Semantic Scene Completion (SSC), which explores taking full advantage of light-weight camera inputs without resorting to any external 3D sensors. Our key insight is to leverage stereo matching to resolve geometric ambiguity. To improve its robustness in unmatched areas, we introduce bird's-eye-view (BEV) representation to inspire hallucination ability with rich context information. On top of the stereo and BEV representations, a mutual interactive aggregation (MIA) module is carefully devised to fully unleash their power. Specifically, a Bi-directional Interaction Transformer (BIT) augmented with confidence re-weighting is used to encourage reliable prediction through mutual guidance while a Dual Volume Aggregation (DVA) module is designed to facilitate complementary aggregation. Experimental results on SemanticKITTI demonstrate that the proposed StereoScene outperforms the state-of-the-art camera-based methods by a large margin with a relative improvement of 26.9% in geometry and 38.6% in semantic.

A Simple Baseline for Supervised Surround-view Depth Estimation

Depth estimation has been widely studied and serves as the fundamental step of 3D perception for autonomous driving. Though significant progress has been made for monocular depth estimation in the past decades, these attempts are mainly conducted on the KITTI benchmark with only front-view cameras, which ignores the correlations across surround-view cameras. In this paper, we propose S3Depth, a Simple Baseline for Supervised Surround-view Depth Estimation, to jointly predict the depth maps across multiple surrounding cameras. Specifically, we employ a global-to-local feature extraction module which combines CNN with transformer layers for enriched representations. Further, the Adjacent-view Attention mechanism is proposed to enable the intra-view and inter-view feature propagation. The former is achieved by the self-attention module within each view, while the latter is realized by the adjacent attention module, which computes the attention across multi-cameras to exchange the multi-scale representations across surround-view feature maps. Extensive experiments show that our method achieves superior performance over existing state-of-the-art methods on both DDAD and nuScenes datasets.

OpenOccupancy: A Large Scale Benchmark for Surrounding Semantic Occupancy Perception

Semantic occupancy perception is essential for autonomous driving, as automated vehicles require a fine-grained perception of the 3D urban structures. However, existing relevant benchmarks lack diversity in urban scenes, and they only evaluate front-view predictions. Towards a comprehensive benchmarking of surrounding perception algorithms, we propose OpenOccupancy, which is the first surrounding semantic occupancy perception benchmark. In the OpenOccupancy benchmark, we extend the large-scale nuScenes dataset with dense semantic occupancy annotations. Previous annotations rely on LiDAR points superimposition, where some occupancy labels are missed due to sparse LiDAR channels. To mitigate the problem, we introduce the Augmenting And Purifying (AAP) pipeline to ~2x densify the annotations, where ~4000 human hours are involved in the labeling process. Besides, camera-based, LiDAR-based and multi-modal baselines are established for the OpenOccupancy benchmark. Furthermore, considering the complexity of surrounding occupancy perception lies in the computational burden of high-resolution 3D predictions, we propose the Cascade Occupancy Network (CONet) to refine the coarse prediction, which relatively enhances the performance by ~30% than the baseline. We hope the OpenOccupancy benchmark will boost the development of surrounding occupancy perception algorithms.

Tri-Perspective View for Vision-Based 3D Semantic Occupancy Prediction

Modern methods for vision-centric autonomous driving perception widely adopt the bird's-eye-view (BEV) representation to describe a 3D scene. Despite its better efficiency than voxel representation, it has difficulty describing the fine-grained 3D structure of a scene with a single plane. To address this, we propose a tri-perspective view (TPV) representation which accompanies BEV with two additional perpendicular planes. We model each point in the 3D space by summing its projected features on the three planes. To lift image features to the 3D TPV space, we further propose a transformer-based TPV encoder (TPVFormer) to obtain the TPV features effectively. We employ the attention mechanism to aggregate the image features corresponding to each query in each TPV plane. Experiments show that our model trained with sparse supervision effectively predicts the semantic occupancy for all voxels. We demonstrate for the first time that using only camera inputs can achieve comparable performance with LiDAR-based methods on the LiDAR segmentation task on nuScenes. Code: https://github.com/wzzheng/TPVFormer.