PVGNet: A Bottom-Up One-Stage 3D Object Detector with Integrated Multi-Level Features

Miao, Zhenwei; Chen, Jikai; Pan, Hongyu; Zhang, Ruiwen; Liu, Kaixuan; Hao, Peihan; Zhu, Jun; Wang, Yan; Zhan, Xin

doi:10.1109/cvpr46437.2021.00329

Cited by 30 publications

(6 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…8) Confidence Score: Confidence score indicates the optimum threshold to categorize false positives to ensure the predicted bounding box contains minimum standard score and often used for model performance evaluation [19,79,106]. Non optimal settings for any proposed model requires more minimized confidence score for precise bounding box detection for 3D object detection.…”

Section: ) Average Orientation Similarity (Aos): Averagementioning

confidence: 99%

Vision based 3D Object Detection using Deep Learning: Methods with Challenges and Applications towards Future Directions

Saif¹,

Mahayuddin²

2022

IJACSA

View full text Add to dashboard Cite

For autonomous intelligent systems, 3D object detection can act as a basis for decision making by providing information such as object's size, position and direction to perceive information about surrounding environment. Successful application using robust 3D object detection can hugely impact robotic industry, augmented and virtual reality sectors in the context of Fourth Industrial Revolution (IR4.0). Recently, deep learning has become potential approach for 3D object detection to learn powerful semantic object features for various tasks, i.e., depth map construction, segmentation and classification. As a result, exponential development in the growth of potential methods is observed in recent years. Although, good number of potential efforts have been made to address 3D object detection, a depth and critical review from different viewpoints is still lacking. As a result, comparison among various methods remains challenging which is important to select method for particular application. Based on strong heterogeneity in previous methods, this research aims to alleviate, analyze and systematize related existing research based on challenges and methodologies from different viewpoints to guide future development and evaluation by bridging the gaps using various sensors, i.e., cameras, LiDAR and Pseudo-LiDAR. At first, this research illustrates critical analysis on existing sophisticated methods by identifying six significant key areas based on current scenarios, challenges, and significant problems to be addressed for solution. Next, this research presents strict comprehensive analysis for validating 3D object detection methods based on eight authoritative 3D detection benchmark datasets depending on the size of the datasets and eight validation matrices. Finally, valuable insights of existing challenges are presented for future directions. Overall extensive review proposed in this research can contribute significantly to embark further investigation in multimodal 3D object detection.

show abstract

Section: ) Average Orientation Similarity (Aos): Averagementioning

confidence: 99%

Vision based 3D Object Detection using Deep Learning: Methods with Challenges and Applications towards Future Directions

Saif¹,

Mahayuddin²

2022

IJACSA

View full text Add to dashboard Cite

show abstract

“…We adopt the voxel-based CNN as the backbone due to its efficiency. In order to prevent the loss of geometry information, which is crucial for implicit boundary generation, we simultaneously extract point-wise and voxelwise features in one backbone [23,48]. As the yellow block shown in Fig.…”

Section: Backbone Networkmentioning

confidence: 99%

“…setting as in KITTI benchmark. In particular, we focus on the 'Car' category as many recent works [6,23,46] and adopt IoU = 0.7 for evaluation. When performing experimental studies on the val set, we use the train data for training.…”

Section: Dataset and Protocolsmentioning

confidence: 99%

ImpDet: Exploring Implicit Fields for 3D Object Detection

Qian¹,

Wang²,

Zhu³

et al. 2022

Preprint

View full text Add to dashboard Cite

Conventional 3D object detection approaches concentrate on bounding boxes representation learning with several parameters, i.e., localization, dimension, and orientation. Despite its popularity and universality, such a straightforward paradigm is sensitive to slight numerical deviations, especially in localization. By exploiting the property that point clouds are naturally captured on the surface of objects along with accurate location and intensity information, we introduce a new perspective that views bounding box regression as an implicit function. This leads to our proposed framework, termed Implicit Detection or ImpDet, which leverages implicit field learning for 3D object detection. Our ImpDet assigns specific values to points in different local 3D spaces, thereby high-quality boundaries can be generated by classifying points inside or outside the boundary. To solve the problem of sparsity on the object surface, we further present a simple yet efficient virtual sampling strategy to not only fill the empty region, but also learn rich semantic features to help refine the boundaries. Extensive experimental results on KITTI and Waymo benchmarks demonstrate the effectiveness and robustness of unifying implicit fields into object detection.

show abstract

“…Instead of using voxels, PointPillars [15] introduced a column (pillar) representation to further reduce the complexity. Hybrid methods such as PV-RCNN [34,35] or STD [53] simultaneously process point and voxel information to obtain multi-scale features and fine-grained localization [12,21,22,54]. For example, Part-A 2 [37] uses pointwise features to predict intra-object parts and then aggregates the part information for box refinement to improve the robustness.…”

Section: Related Workmentioning

confidence: 99%

OccAM's Laser: Occlusion-based Attribution Maps for 3D Object Detectors on LiDAR Data

Schinagl¹,

Krispel²,

Possegger³

et al. 2022

Preprint

View full text Add to dashboard Cite

While 3D object detection in LiDAR point clouds is wellestablished in academia and industry, the explainability of these models is a largely unexplored field. In this paper, we propose a method to generate attribution maps for the detected objects in order to better understand the behavior of such models. These maps indicate the importance of each 3D point in predicting the specific objects. Our method works with black-box models: We do not require any prior knowledge of the architecture nor access to the model's internals, like parameters, activations or gradients. Our efficient perturbation-based approach empirically estimates the importance of each point by testing the model with randomly generated subsets of the input point cloud. Our sub-sampling strategy takes into account the special characteristics of LiDAR data, such as the depth-dependent point density. We show a detailed evaluation of the attribution maps and demonstrate that they are interpretable and highly informative. Furthermore, we compare the attribution maps of recent 3D object detection architectures to provide insights into their decision-making processes. Car Car Pedestrian (Wall) Car Cyclist Car

show abstract

PVGNet: A Bottom-Up One-Stage 3D Object Detector with Integrated Multi-Level Features

Cited by 30 publications

References 25 publications

Vision based 3D Object Detection using Deep Learning: Methods with Challenges and Applications towards Future Directions

Vision based 3D Object Detection using Deep Learning: Methods with Challenges and Applications towards Future Directions

ImpDet: Exploring Implicit Fields for 3D Object Detection

OccAM's Laser: Occlusion-based Attribution Maps for 3D Object Detectors on LiDAR Data

Contact Info

Product

Resources

About