PCT: Point cloud transformer

Guo, Minghao; Cai, Jin-Yi; Liu, Zheng-Ning; Mu, Tai‐Jiang; Martin, Ralph R.; Hu, Shi‐Min

doi:10.1007/s41095-021-0229-5

Cited by 951 publications

(399 citation statements)

References 20 publications

(26 reference statements)

Supporting

Mentioning

296

Contrasting

Order By: Relevance

“…whereas LBR represents three operations: linear operation, batch processing, and the ReLU activation function. F a = A(Q, K, V) and F in − F a are similar to a discrete Laplace operator [29]. Our experiments showed that self-attention was more beneficial for subsequent registration tasks when replaced by offset-attention.…”

Section: Transformermentioning

confidence: 73%

“…Attention mechanisms use relative importance to focus on different parts of the input sequence, highlighting the relationship between inputs enabling the capture of context and high-order dependencies. Inspired by the point cloud processing Transformer proposed in recent years [29,30], we define Q, K, and V as the query, key, and value matrices, respectively, generated by the linear transformation of input characteristics.𝐹 𝑖𝑛 ∈ ℝ 𝑑 The function () A  describes the mapping of N queries 𝑄 ∈ ℝ 𝑁×𝑑 𝑘 and N key values to 𝐾 ∈ ℝ 𝑁 𝑘 ×𝑑 𝑘 and 𝑉 ∈ ℝ 𝑁 𝑘 ×𝑑𝑣 to the output [31]. The attention weight is calculated by the matrix dot product 𝑄𝐾 𝑇 ∈ ℝ 𝑁×𝑑 :…”

Section: Transformermentioning

confidence: 99%

“…Our graph convolution network research studied the benefits of using a Laplace matrix [32] L = D − E instead of an adjacency matrix E, where D is the antiangular matrix. Using the former, the output characteristics of the offset-attention used by PCTNet [29] are…”

Section: Transformermentioning

confidence: 99%

See 2 more Smart Citations

PTRNet: Global Feature and Local Feature Encoding for Point Cloud Registration

Yang

Shi

et al. 2022

Applied Sciences

View full text Add to dashboard Cite

Existing end-to-end cloud registration methods are often inefficient and susceptible to noise. We propose an end-to-end point cloud registration network model, Point Transformer for Registration Network (PTRNet), that considers local and global features to improve this behavior. Our model uses point clouds as inputs and applies a Transformer method to extract their global features. Using a K-Nearest Neighbor (K-NN) topology, our method then encodes the local features of a point cloud and integrates them with the global features to obtain the point cloud’s strong global features. Comparative experiments using the ModelNet40 data set show that our method offers better results than other methods, with a mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) between the ground truth and predicted values lower than those of competing methods. In the case of multi-object class without noise, the rotation average absolute error of PTRNet is reduced to 1.601 degrees and the translation average absolute error is reduced to 0.005 units. Compared to other recent end-to-end registration methods and traditional point cloud registration methods, the PTRNet method has less error, higher registration accuracy, and better robustness.

show abstract

Section: Transformermentioning

confidence: 73%

Section: Transformermentioning

confidence: 99%

See 1 more Smart Citation

PTRNet: Global Feature and Local Feature Encoding for Point Cloud Registration

Yang

Shi

et al. 2022

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…CN [50] proposes a novel channel normalization scheme to balance information from different layers to benefit the model in integrating multilayer structure information. PCT [51] is based on Transformer, and the fundamental idea is introducing the inherent order invariance of Transformer and learning feature through the attention mechanism.…”

Section: Point-based Methodsmentioning

confidence: 99%

A Lightweight Structure Based on Feature Fusion for Point Cloud Analysis

Zheng

Sun

2021

IEEE Access

View full text Add to dashboard Cite

Point cloud analysis is challenging due to the irregularity and sparsity, making it difficult to capture the underlying geometric characteristics. This paper proposes a lightweight structure that effectively aggregates the local patterns and the spatial layout of them extracted from the irregular and sparse point cloud. Unlike previous works that seek sophisticated feature extraction methods, the key to this structure is simultaneously exploring the local features and their distribution features in a concise manner. Specifically, the two features, which correspond to two different scales, are extracted independently and then fused. In this way, an abstract shape-level feature that contains much shape awareness and robustness is obtained. Moreover, since the structure exhibits rapid convergence with conventional learning schedule, we upgrade it by introducing the snapshot ensemble and creatively design a more flexible and effective cyclic annealing learning schedule. We evaluate the structure on challenging benchmarks, and experiment results prove that our model achieves on-par or better performance than previous state-of-the-art (SOTA) methods, although with the simple shared multi-layer perceptrons (MLPs) as feature extractors.

show abstract

“…This mechanism is actually well suited to dealing with data like point clouds. PCT [31] enhances input embedding by supporting farthest point sampling and nearest neighbor search. It applies transformer to point clouds and achieves good results.…”

Section: Introductionmentioning

confidence: 99%

Radar Transformer: An Object Classification Network Based on 4D MMW Imaging Radar

Bai

Zheng

et al. 2021

Sensors

View full text Add to dashboard Cite

Automotive millimeter-wave (MMW) radar is essential in autonomous vehicles due to its robustness in all weather conditions. Traditional commercial automotive radars are limited by their resolution, which makes the object classification task difficult. Thus, the concept of a new generation of four-dimensional (4D) imaging radar was proposed. It has high azimuth and elevation resolution and contains Doppler information to produce a high-quality point cloud. In this paper, we propose an object classification network named Radar Transformer. The algorithm takes the attention mechanism as the core and adopts the combination of vector attention and scalar attention to make full use of the spatial information, Doppler information, and reflection intensity information of the radar point cloud to realize the deep fusion of local attention features and global attention features. We generated an imaging radar classification dataset and completed manual annotation. The experimental results show that our proposed method achieved an overall classification accuracy of 94.9%, which is more suitable for processing radar point clouds than the popular deep learning frameworks and shows promising performance.

show abstract

PCT: Point cloud transformer

Cited by 951 publications

References 20 publications

PTRNet: Global Feature and Local Feature Encoding for Point Cloud Registration

PTRNet: Global Feature and Local Feature Encoding for Point Cloud Registration

A Lightweight Structure Based on Feature Fusion for Point Cloud Analysis

Radar Transformer: An Object Classification Network Based on 4D MMW Imaging Radar

Contact Info

Product

Resources

About