Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MinkUNeXt: Point Cloud-based Large-scale Place Recognition using 3D Sparse Convolutions

J. J. Cabrera, A. Santo, A. Gil, C. Viegas, L. Payá

Abstract

This paper presents MinkUNeXt, an effective and efficient architecture for place-recognition from point clouds entirely based on the new 3D MinkNeXt Block, a residual block composed of 3D sparse convolutions that follows the philosophy established by recent Transformers but purely using simple 3D convolutions. Feature extraction is performed at different scales by a U-Net encoder-decoder network and the feature aggregation of those features into a single descriptor is carried out by a Generalized Mean Pooling (GeM). The proposed architecture demonstrates that it is possible to surpass the current state-of-the-art by only relying on conventional 3D sparse convolutions without making use of more complex and sophisticated proposals such as Transformers, Attention-Layers or Deformable Convolutions. A thorough assessment of the proposal has been carried out using the Oxford RobotCar and the In-house datasets. As a result, MinkUNeXt proves to outperform other methods in the state-of-the-art.

MinkUNeXt: Point Cloud-based Large-scale Place Recognition using 3D Sparse Convolutions

Abstract

This paper presents MinkUNeXt, an effective and efficient architecture for place-recognition from point clouds entirely based on the new 3D MinkNeXt Block, a residual block composed of 3D sparse convolutions that follows the philosophy established by recent Transformers but purely using simple 3D convolutions. Feature extraction is performed at different scales by a U-Net encoder-decoder network and the feature aggregation of those features into a single descriptor is carried out by a Generalized Mean Pooling (GeM). The proposed architecture demonstrates that it is possible to surpass the current state-of-the-art by only relying on conventional 3D sparse convolutions without making use of more complex and sophisticated proposals such as Transformers, Attention-Layers or Deformable Convolutions. A thorough assessment of the proposal has been carried out using the Oxford RobotCar and the In-house datasets. As a result, MinkUNeXt proves to outperform other methods in the state-of-the-art.
Paper Structure (17 sections, 2 equations, 4 figures, 5 tables)

This paper contains 17 sections, 2 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. State of the art
  3. MinkUNeXt: global point cloud descriptor for place recognition
  4. Global Architecture
  5. Residual Block Architecture
  6. Experiments
  7. Datasets
  8. Labelling and similarity
  9. Training and evaluation
  10. Implementation details
  11. Ablation study: From MinkUNet to MinkUNeXt
  12. Global Design
  13. Residual Block Design
  14. Comparison with the state of the art
  15. Results with the Baseline Protocol
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Point cloud-based place recognition. Each query point cloud (red) is embedded into a global descriptor which is compared with the descriptors from the database point clouds (blue) by means of a Nearest Neighbour Search.
  • Figure 2: This diagram shows the architecture of the proposed MinkUNeXt, which is based on a semantic segmentation network (U-Net) modified and enhanced to perform place-recognition from point clouds.
  • Figure 3: This diagram shows the proposed MinkNeXt Block. This residual block is an essential part of the global network, since increases the number of feature maps through an inverted bottleneck.
  • Figure 4: This diagram illustrates the design progress of the proposed architecture from MinkUNet up to MinkUNeXt. All the proposed modifications are summarized in Table \ref{['tab:comentarios_nodo']}.