TransLoc3D : Point Cloud based Large-scale Place Recognition using Adaptive Receptive Fields

Xu, Tianxing; Guo, Yuan-Chen; Lai, Yu‐Kun; Zhang, Song-Hai

doi:10.48550/arxiv.2105.11605

Cited by 3 publications

(10 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The availability of the efficient 3D sparse convolution library sparked an interest in using sparse volumetric representation for place recognition purposes [8], [9], [11]. The first method, MinkLoc3D [8], surpasses previous methods by a significant margin when evaluated on the Oxford RobotCar dataset, proving that the data representation is a critical component of a 3D LiDAR place recognition method.…”

Section: B 3d Lidar Place Recognitionmentioning

confidence: 99%

“…MinkLoc++ [9] further improves MinkLoc3D with a channel attention mechanism ECA [35] while fusing 3D LiDAR scans and camera images. Similar approach is used in TransLoc3D [11], which combines sparse convolutions, adaptive receptive field module (ARFM) based on ECA, and Transformer module.…”

Section: B 3d Lidar Place Recognitionmentioning

confidence: 99%

“…Oxford RobotCar Source of results AR@1% AR@1 PointNetVLAD [6] [6] 80.3 63.3 Scan Context [4] [31] 81.9 64.6 PCAN [12] [12] 83.8 70.7 DH3D-4096 [14] [14] 84.3 73.3 EPC-Net [16] [16] 94.7 86.2 LPD-Net [13] [13] 94.9 86.4 DISCO [31] [31] 75.0 88.4 SOE-Net [17] [17] 96.4 89.3 NDT-Transformer [7] [7] 97.7 93.8 TransLoc3D [11] [11] 98.5 95.0 MinkLoc3D (3D) [8] [8] 97.9 93.8 MinkLoc3D-S (our) our evaluation 93.1 82.0…”

Section: Table IIImentioning

confidence: 99%

“…we have already explored image-like [1], voxel-based [2], bird's-eye view approaches [3], [4], [5], or unordered point sets [6], [7] representations. More recently, we also see a rise in sparse convolution-based approaches [8], [9] and attentionbased modules [10] that might be combined together [11] as well. Many of these methods are trained, evaluated, and compared on the Oxford RobotCar dataset [6], [8], [12], [13], [14], [15], [16], [17], which was gathered by concatenating multiple 2D LiDAR scans covering a 20 m distance and subsampling to 4096 points [6].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

MinkLoc3D-SI: 3D LiDAR Place Recognition With Sparse Convolutions, Spherical Coordinates, and Intensity

Kamil

Banaszczyk

Nowicki

et al. 2022

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

The 3D LiDAR place recognition aims to estimate a coarse localization in a previously seen environment based on a single scan from a rotating 3D LiDAR sensor. The existing solutions to this problem include hand-crafted point cloud descriptors (e.g., ScanContext, M2DP, LiDAR IRIS) and deep learning-based solutions (e.g., PointNetVLAD, PCAN, LPD-Net, DAGC, MinkLoc3D), which are often only evaluated on accumulated 2D scans from the Oxford RobotCar dataset. We introduce MinkLoc3D-SI, a sparse convolution-based solution that utilizes spherical coordinates of 3D points and processes the intensity of 3D LiDAR measurements, improving the performance when a single 3D LiDAR scan is used. Our method integrates the improvements typical for hand-crafted descriptors (like ScanContext) with the most efficient 3D sparse convolutions (MinkLoc3D). Our experiments show improved results on single scans from 3D LiDARs (USyd Campus dataset) and great generalization ability (KITTI dataset). Using intensity information on accumulated 2D scans (RobotCar Intensity dataset) improves the performance, even though spherical representation doesn't produce a noticeable improvement. As a result, MinkLoc3D-SI is suited for single scans obtained from a 3D LiDAR, making it applicable in autonomous vehicles.

show abstract

Section: B 3d Lidar Place Recognitionmentioning

confidence: 99%

Section: B 3d Lidar Place Recognitionmentioning

confidence: 99%

Section: Table IIImentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

MinkLoc3D-SI: 3D LiDAR Place Recognition With Sparse Convolutions, Spherical Coordinates, and Intensity

Kamil

Banaszczyk

Nowicki

et al. 2022

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

“…There's a long line of research on point cloud-based place recognition using learned descriptors [1]- [8], with a progressive advancements in state of the art. PointNet-like architecture used in early works [1]- [3] was not well suited to extract informative features and was replaced by a 3D convolutional network based on a sparse voxelized representation in [4].…”

Section: Introductionmentioning

confidence: 99%

Improving Point Cloud Based Place Recognition with Ranking-based Loss and Large Batch Training

Komorowski¹

2022

Preprint

View full text Add to dashboard Cite

The paper presents a simple and effective learningbased method for computing a discriminative 3D point cloud descriptor for place recognition purposes. Recent state-of-theart methods have relatively complex architectures such as multiscale pyramid of point Transformers combined with a pyramid of feature aggregation modules. Our method uses a simple and efficient 3D convolutional feature extraction, based on a sparse voxelized representation, enhanced with channel attention blocks. We employ recent advances in image retrieval and propose a modified version of a loss function based on a differentiable average precision approximation. Such loss function requires training with very large batches for the best results. This is enabled by using multistaged backpropagation. Experimental evaluation on the popular benchmarks proves the effectiveness of our approach, with a consistent improvement over state of the art.

show abstract

A Single Stage and Single View 3D Point Cloud Reconstruction Network Based on DetNet

Zhu

2022

Sensors

View full text Add to dashboard Cite

It is a challenging problem to infer objects with reasonable shapes and appearance from a single picture. Existing research often pays more attention to the structure of the point cloud generation network, while ignoring the feature extraction of 2D images and reducing the loss in the process of feature propagation in the network. In this paper, a single-stage and single-view 3D point cloud reconstruction network, 3D-SSRecNet, is proposed. The proposed 3D-SSRecNet is a simple single-stage network composed of a 2D image feature extraction network and a point cloud prediction network. The single-stage network structure can reduce the loss of the extracted 2D image features. The 2D image feature extraction network takes DetNet as the backbone. DetNet can extract more details from 2D images. In order to generate point clouds with better shape and appearance, in the point cloud prediction network, the exponential linear unit (ELU) is used as the activation function, and the joint function of chamfer distance (CD) and Earth mover’s distance (EMD) is used as the loss function of 3DSSRecNet. In order to verify the effectiveness of 3D-SSRecNet, we conducted a series of experiments on ShapeNet and Pix3D datasets. The experimental results measured by CD and EMD have shown that 3D-SSRecNet outperforms the state-of-the-art reconstruction methods.

show abstract

TransLoc3D : Point Cloud based Large-scale Place Recognition using Adaptive Receptive Fields

Cited by 3 publications

References 56 publications

MinkLoc3D-SI: 3D LiDAR Place Recognition With Sparse Convolutions, Spherical Coordinates, and Intensity

MinkLoc3D-SI: 3D LiDAR Place Recognition With Sparse Convolutions, Spherical Coordinates, and Intensity

Improving Point Cloud Based Place Recognition with Ranking-based Loss and Large Batch Training

A Single Stage and Single View 3D Point Cloud Reconstruction Network Based on DetNet

Contact Info

Product

Resources

About