GPr-Net: Geometric Prototypical Network for Point Cloud Few-Shot Learning

12 Apr 2023  ·  Tejas Anvekar, Dena Bazazian ·

In the realm of 3D-computer vision applications, point cloud few-shot learning plays a critical role. However, it poses an arduous challenge due to the sparsity, irregularity, and unordered nature of the data. Current methods rely on complex local geometric extraction techniques such as convolution, graph, and attention mechanisms, along with extensive data-driven pre-training tasks. These approaches contradict the fundamental goal of few-shot learning, which is to facilitate efficient learning. To address this issue, we propose GPr-Net (Geometric Prototypical Network), a lightweight and computationally efficient geometric prototypical network that captures the intrinsic topology of point clouds and achieves superior performance. Our proposed method, IGI++ (Intrinsic Geometry Interpreter++) employs vector-based hand-crafted intrinsic geometry interpreters and Laplace vectors to extract and evaluate point cloud morphology, resulting in improved representations for FSL (Few-Shot Learning). Additionally, Laplace vectors enable the extraction of valuable features from point clouds with fewer points. To tackle the distribution drift challenge in few-shot metric learning, we leverage hyperbolic space and demonstrate that our approach handles intra and inter-class variance better than existing point cloud few-shot learning methods. Experimental results on the ModelNet40 dataset show that GPr-Net outperforms state-of-the-art methods in few-shot learning on point clouds, achieving utmost computational efficiency that is $170\times$ better than all existing works. The code is publicly available at https://github.com/TejasAnvekar/GPr-Net.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (10-shot) GPr-Net + Hyp (1024) Overall Accuracy 70.4 # 19
Standard Deviation 1.8 # 3
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (10-shot) GPr-Net + Hyp (512) Overall Accuracy 71.6 # 18
Standard Deviation 1.1 # 1
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (10-shot) GPr-Net + Euc (1024) Overall Accuracy 62.1 # 21
Standard Deviation 1.9 # 5
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (10-shot) GPr-Net + Euc (512) Overall Accuracy 62.3 # 20
Standard Deviation 2.0 # 6
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (20-shot) GPr-Net + Hyp (1024) Overall Accuracy 72.8 # 19
Standard Deviation 1.8 # 3
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (20-shot) GPr-Net + Euc (1024) Overall Accuracy 63.4 # 20
Standard Deviation 2.0 # 4
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (20-shot) GPr-Net + Hyp (512) Overall Accuracy 73.8 # 18
Standard Deviation 2.0 # 4
Few-Shot 3D Point Cloud Classification ModelNet40 10-way (20-shot) GPr-Net + Euc (512) Overall Accuracy 63.3 # 21
Standard Deviation 2.2 # 6
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (10-shot) GPr-Net + Euc (1024) Overall Accuracy 74.4 # 19
Standard Deviation 2.0 # 9
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (10-shot) GPr-Net + Hyp (1024) Overall Accuracy 80.4 # 18
Standard Deviation 0.5 # 1
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (10-shot) GPr-Net + Euc (512) Overall Accuracy 74.0 # 20
Standard Deviation 2.3 # 11
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (10-shot) GPr-Net + Hyp (512) Overall Accuracy 81.1 # 17
Standard Deviation 1.5 # 2
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (20-shot) GPr-Net + Euc (1024) Overall Accuracy 75.1 # 19
Standard Deviation 2.1 # 17
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (20-shot) GPr-Net + Euc (512) Overall Accuracy 75.0 # 20
Standard Deviation 2.4 # 18
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (20-shot) GPr-Net + Hyp (512) Overall Accuracy 82.7 # 17
Standard Deviation 1.3 # 6
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (20-shot) GPr-Net + Hyp (1024) Overall Accuracy 82.0 # 18
Standard Deviation 0.9 # 2

Methods


No methods listed for this paper. Add relevant methods here