Accurate 3D object detection from images can be hindered by inherent depth ambiguity. While knowledge distillation (KD) from privileged sensors such as LiDAR offers a promising direction, it often suffers from a critical cross-sensor domain gap. To address this, we introduce DK3D, a novel depth-aware knowledge distillation framework for 3D detection. Our core strategy involves providing the teacher with privileged ground-truth depth during training. This directly avoids the feature representation mismatch and subsequent inefficient knowledge transfer required when distilling from a LiDAR teacher (sparse, geometric) to a camera-based student (dense, semantic). DK3D introduces specialized modules tailored for two primary student paradigms. For depth-assisted models, we employ a channel-wise projection layer (CPL) and an adversarial scoring block (ASB) to align intermediate features at both the pixel and distribution levels. For depth-independent models, a novel vision-depth association module allows the student to implicitly reason about geometry by fusing depth cues with visual features. Both approaches are further enhanced by target-aware spatial response distillation, which captures complex inter-object spatial relationships. Extensive experiments on the KITTI and nuScenes benchmarks demonstrate that DK3D significantly improves performance for both monocular and multi-view 3D detection, outperforming state-of-the-art methods. As a versatile, plug-and-play framework, DK3D boosts existing models without requiring additional training data or increasing the computational cost at inference.

Monocular 3D object detection usually adopts direct or hierarchical label supervision. Recently, the distillation supervision transfers the spatial knowledge from LiDAR- or stereo-based teacher networks to monocular detectors, but remaining the domain gap. To mitigate this issue and pursue adequate label manipulation, we exploit Foreground Depth map for feature-supervised monocular 3D object detection named FD3D, which develops the high-quality instructive intermediate features to conduct desirable auxiliary feature supervision with only the original image and annotation foreground object-wise depth map (AFOD) as input. Furthermore, we build up our instructive feature generation network to create instructive spatial features based on the sufficient correlation between image features and pre-processed AFOD, where AFOD provides the attention focus only on foreground objects to achieve clearer guidance in the detection task. Moreover, we apply the auxiliary feature supervision from the pixel and distribution level to achieve comprehensive spatial knowledge guidance. Extensive experiments demonstrate that our method achieves state-of-the-art performance on both the KITTI and nuScenes datasets, with no external data and no extra inference computational cost. We also conduct quantitative and qualitative studies to reveal the effectiveness of our designs.

Radar-camera 3D object detection aims at interacting radar signals with camera images for identifying objects of interest and localizing their corresponding 3D bounding boxes. To overcome the severe sparsity and ambiguity of radar signals, we propose a robust framework based on probabilistic denoising diffusion modeling. We design our framework to be easily implementable on different multi-view 3D detectors without the requirement of using LiDAR point clouds during either the training or inference. In specific, we first design our framework with a denoised radar-camera encoder via developing a lightweight denoising diffusion model with semantic embedding. Secondly, we develop the query denoising training into 3D space via introducing the reconstruction training at depth measurement for the transformer detection decoder. Our framework achieves new state-of-the-art performance on the nuScenes 3D detection benchmark but with few computational cost increases compared to the baseline detectors.

Monocular 3D object detection is a low-cost but challenging task, as it requires generating accurate 3D localization solely from a single image input. Recent developed depth-assisted methods show promising results by using explicit depth maps as intermediate features, which are either precomputed by monocular depth estimation networks or jointly evaluated with 3D object detection. However, inevitable errors from estimated depth priors may lead to misaligned semantic information and 3D localization, hence resulting in feature smearing and suboptimal predictions. To mitigate this issue, we propose ADD, an Attention-based Depth knowledge Distillation framework with 3D-aware positional encoding. Unlike previous knowledge distillation frameworks that adopt stereo- or LiDAR-based teachers, we build up our teacher with identical architecture as the student but with extra ground-truth depth as input. Credit to our teacher design, our framework is seamless, domain-gap free, easily implementable, and is compatible with object-wise ground-truth depth. Specifically, we leverage intermediate features and responses for knowledge distillation. Considering long-range 3D dependencies, we propose 3D-aware self-attention and target-aware cross-attention modules for student adaptation. Extensive experiments are performed to verify the effectiveness of our framework on the challenging KITTI 3D object detection benchmark. We implement our framework on three representative monocular detectors, and we achieve state-of-the-art performance with no additional inference computational cost relative to baseline models. Our code is available at https://github.com/rockywind/ADD.

Multi-view radar-camera fused 3D object detection provides a farther detection range and more helpful features for autonomous driving, especially under adverse weather. The current radar-camera fusion methods deliver kinds of designs to fuse radar information with camera data. However, these fusion approaches usually adopt the straightforward concatenation operation between multi-modal features, which ignores the semantic alignment with radar features and sufficient correlations across modals. In this paper, we present MVFusion, a novel Multi-View radar-camera Fusion method to achieve semantic-aligned radar features and enhance the cross-modal information interaction. To achieve so, we inject the semantic alignment into the radar features via the semantic-aligned radar encoder (SARE) to produce image-guided radar features. Then, we propose the radar-guided fusion transformer (RGFT) to fuse our radar and image features to strengthen the two modals' correlation from the global scope via the cross-attention mechanism. Extensive experiments show that MVFusion achieves state-of-the-art performance (51.7% NDS and 45.3% mAP) on the nuScenes dataset. We shall release our code and trained networks upon publication.

Monocular 3D object detection reveals an economical but challenging task in autonomous driving. Recently center-based monocular methods have developed rapidly with a great trade-off between speed and accuracy, where they usually depend on the object center's depth estimation via 2D features. However, the visual semantic features without sufficient pixel geometry information, may affect the performance of clues for spatial 3D detection tasks. To alleviate this, we propose MonoPGC, a novel end-to-end Monocular 3D object detection framework with rich Pixel Geometry Contexts. We introduce the pixel depth estimation as our auxiliary task and design depth cross-attention pyramid module (DCPM) to inject local and global depth geometry knowledge into visual features. In addition, we present the depth-space-aware transformer (DSAT) to integrate 3D space position and depth-aware features efficiently. Besides, we design a novel depth-gradient positional encoding (DGPE) to bring more distinct pixel geometry contexts into the transformer for better object detection. Extensive experiments demonstrate that our method achieves the state-of-the-art performance on the KITTI dataset.

The monocular depth estimation task has recently revealed encouraging prospects, especially for the autonomous driving task. To tackle the ill-posed problem of 3D geometric reasoning from 2D monocular images, multi-frame monocular methods are developed to leverage the perspective correlation information from sequential temporal frames. However, moving objects such as cars and trains usually violate the static scene assumption, leading to feature inconsistency deviation and misaligned cost values, which would mislead the optimization algorithm. In this work, we present CTA-Depth, a Context-aware Temporal Attention guided network for multi-frame monocular Depth estimation. Specifically, we first apply a multi-level attention enhancement module to integrate multi-level image features to obtain an initial depth and pose estimation. Then the proposed CTA-Refiner is adopted to alternatively optimize the depth and pose. During the CTA-Refiner process, context-aware temporal attention (CTA) is developed to capture the global temporal-context correlations to maintain the feature consistency and estimation integrity of moving objects. In particular, we propose a long-range geometry embedding (LGE) module to produce a long-range temporal geometry prior. Our approach achieves significant improvements (e.g., 13.5% for the Abs Rel metric on the KITTI dataset) over state-of-the-art approaches on three benchmark datasets.

Due to the lack of spatial cues giving merely a single image, many monocular depth estimation methods have been developed to leverage stereo or multi-view images to learn the spatial information of a scene in a self-supervised manner. However, these methods have limited performance gain since they are not able to exploit sufficient 3D geometry cues during inference, where only monocular images are available. In this work, we present V2Depth, a novel coarse-to-fine framework with Virtual View feature simulation for supervised monocular Depth estimation. Specifically, we first design a virtual-view feature simulator by leveraging the technique of novel view synthesis and contrastive learning to generate virtual view feature maps. In this way, we explicitly provide representative spatial geometry for subsequent depth estimation in both the training and inference stages. Then we introduce a 3DVA-Refiner to iteratively optimize the predicted depth map. During the optimization process, 3D-aware virtual attention is developed to capture the global spatial-context correlations to maintain the feature consistency of different views and estimation integrity of the 3D scene such as objects with occlusion relationships. Decisive improvements over state-of-the-art approaches on three benchmark datasets across all metrics demonstrate the superiority of our method.

Abstract Channel and spatial attention mechanisms have proven to provide an evident performance boost of deep convolution neural networks. Most existing methods focus on one or run them parallel (series), neglecting the collaboration between the two attentions. In order to better establish the feature interaction between the two types of attentions, a plug‐and‐play attention module is proposed, which is termed as ‘CAT’—activating the Collaboration between spatial and channel Attentions based on learned Traits. Specifically, traits are represented as trainable coefficients (i.e. colla‐factors) to adaptively combine contributions of different attention modules to fit different image hierarchies and tasks better. Moreover, the global entropy pooling is proposed apart from global average pooling and global maximum pooling (GMP) operators, which is an effective component in suppressing noise signals by measuring the information disorder of feature maps. A three‐way pooling operation is introduced into attention modules and the adaptive mechanism is applied to fuse their outcomes. Extensive experiments on MS COCO, Pascal‐VOC, Cifar‐100, and ImageNet show that our CAT outperforms the existing state‐of‐the‐art attention mechanisms in object detection, instance segmentation, and image classification. The model and code will be released soon.

In recent years, Internet of Things (IoT) has made rapid development, and IoT devices are developing towards intelligence. IoT terminal devices represented by surveillance cameras play an irreplaceable role in modern society, most of them are integrated with video action recognition and other intelligent functions. However, their performance is somewhat affected by the limitation of computing resources of IoT terminal devices and the lack of long-range non-linear temporal relation modeling and reverse motion information modeling. To address this urgent problem, we introduce a startling Temporal Transformer Network with Self-supervision (TTSN). Our high-performance TTSN mainly consists of a temporal transformer module and a temporal sequence self-supervision module. Concisely speaking, we utilize the efficient temporal transformer module to model the non-linear temporal dependencies among non-local frames, which significantly enhances complex motion feature representations. The temporal sequence self-supervision module we employ unprecedentedly adopts the streamlined strategy of “random batch random channel” to reverse the sequence of video frames, allowing robust extractions of motion information representation from inversed temporal dimensions and improving the generalization capability of the model. Extensive experiments on three widely used datasets (HMDB51, UCF101, and Something-something V1) have conclusively demonstrated that our proposed TTSN is promising as it successfully achieves state-of-the-art performance for video action recognition. Our TTSN provides the possibility for its application in IoT scenarios due to its computational complexity and high performance. With the rapid development of the Internet of Things (IoT), more and more data is being disseminated in the form of video, which also puts new requirements on the understanding and modeling of video data. In recent years, 2D Convolutional Networks-based video action recognition has encouragingly gained wide popularity; However, constrained by the lack of long-range non-linear temporal relation modeling and reverse motion information modeling, the performance of existing models is, therefore, undercut seriously. To address this urgent problem, we introduce a startling Temporal Transformer Network with Self-supervision (TTSN). Our high-performance TTSN mainly consists of a temporal transformer module and a temporal sequence self-supervision module. Concisely speaking, we utilize the efficient temporal transformer module to model the non-linear temporal dependencies among non-local frames, which significantly enhances complex motion feature representations. The temporal sequence self-supervision module we employ unprecedentedly adopts the streamlined strategy of “random batch random channel” to reverse the sequence of video frames, allowing robust extractions of motion information representation from inversed temporal dimensions and improving the generalization capability of the model. Extensive experiments on three widely used datasets (HMDB51, UCF101, and Something-something V1) have conclusively demonstrated that our proposed TTSN is promising as it successfully achieves state-of-the-art performance for video action recognition. As a result, our work provides new attention and self-supervised algorithm for processing video data in IoT.

It’s worth noting that the owner-member relationship between wheels and vehicles has an significant contribution to the 3D perception of vehicles, especially in the embedded environment. However, there are currently two main challenges about the above relationship prediction: i) The traditional heuristic methods based on IoU can hardly deal with the traffic jam scenarios for the occlusion. ii) It is difficult to establish an efficient applicable solution for the vehicle-mounted system. To address these issues, we propose an innovative relationship prediction method, namely DeepWORD, by designing a graph convolution network (GCN). Specifically, we utilize the feature maps with local correlation as the input of nodes to improve the information richness. Besides, we introduce the graph attention network (GAT) to dynamically amend the prior estimation deviation. Furthermore, we establish an annotated owner-member relationship dataset called WORD as a large-scale benchmark, which will be available soon. The experiments demonstrate that our solution achieves state-of-the-art accuracy and real-time in practice.

Multi-Label Image Classification (MLIC) aims to predict a set of labels that present in an image. The key to deal with such problem is to mine the associations between image contents and labels, and further obtain the correct assignments between images and their labels. In this paper, we treat each image as a bag of instances, and reformulate the task of MLIC as a instance-label matching selection problem. To model such problem, we propose a novel deep learning framework named Graph Matching based Multi-Label Image Classification (GM-MLIC), where Graph Matching (GM) scheme is introduced owing to its excellent capability of excavating the instance and label relationship. Specifically, we first construct an instance spatial graph and a label semantic graph respectively, and then incorporate them into a constructed assignment graph by connecting each instance to all labels. Subsequently, the graph network block is adopted to aggregate and update all nodes and edges state on the assignment graph to form structured representations for each instance and label. Our network finally derives a prediction score for each instance-label correspondence and optimizes such correspondence with a weighted cross-entropy loss. Extensive experiments conducted on various datasets demonstrate the superiority of our proposed method.

The 3D visual perception for vehicles with the surround-view fisheye camera system is a critical and challenging task for low-cost urban autonomous driving. While existing monocular 3D object detection methods perform not well enough on the fisheye images for mass production, partly due to the lack of 3D datasets of such images. In this paper, we manage to overcome and avoid the difficulty of acquiring the large scale of accurate 3D labeled truth data, by breaking down the 3D object detection task into some sub-tasks, such as vehicle’s contact point detection, type classification, re-identification and unit assembling, etc. Particularly, we propose the concept of Multidimensional Vector to include the utilizable information generated in different dimensions and stages, instead of the descriptive approach for the bird’s eye view (BEV) or a cube of eight points. The experiments of real fisheye images demonstrate that our solution achieves state-of-the-art accuracy while being real-time in practice.

Abstract Generic object detection is a crucial task for autonomous driving. To devise a safe and efficient object detector, the following aspects are required to be considered: high accuracy, real‐time inference speed and small model size. Herein, a simple yet effective anchor‐free object detector named L4Net is proposed, which incorporates a keypoint detection backbone and a co‐attention scheme into a unified framework, and achieves lower computation cost with higher detection accuracy than prior art across a wide spectrum of resource constrains. Specifically, the backbone utilizes M ulti‐scale R eceptive‐fields E nhancement module (MRE) to capture context‐wise information, where the features of object scale and shape invariance are simultaneously considered. The co‐attention scheme integrates the strength of both C lass‐agnostic A ttention (CA) and S emantic A ttention (SA), and explores the valuable features from low‐level to high‐level to generate more accurate prediction boxes. Compared with previous feature fusion strategy, multi‐scale features are selectively integrated by fully exploiting the different characteristics of low‐level and high‐level features, which leads to a small model size and faster inference speed. Extensive experiments on four well‐known datasets demonstrate the effectiveness of our method. For instance, L4Net achieves 71.68% mAP on KITTI test set, with 13.7 M model size at the speed of 149 FPS on NVIDIA TX and 30.7 FPS on Qualcomm‐based device, respectively, which is 4 x smaller and 2 x faster than baseline model.

While real-time parking slot detection plays a critical role in valet parking systems, existing methods have limited success in real-world application. We argue two reasons accounting for the unsatisfactory performance: i, The available datasets have limited diversity, which causes the low generalization ability. ii, Expert knowledge for parking slot detection is under-estimated. Thus, we annotate a large-scale benchmark for training the network and release it for the benefit of community. Driven by the observation of various parking lots in our benchmark, we propose the circular descriptor to regress the coordinates of parking slot vertexes and accordingly localize slots accurately. To further boost the performance, we develop a two-stage deep architecture to localize vertexes in the coarse-to-fine manner. In our benchmark and other datasets, it achieves the state-of-the-art accuracy while being real-time in practice. Benchmark is available at: https://github.com/wuzzh/Parking-slot-dataset