Hyperbolic cosine transformer for LiDAR 3D object detection

Abstract

Previous point-wise methods are suffering from time consumption and limited receptive fields to capture information among points. To address these limitations, we propose the cosh-attention, which reduces the computation complexity of space and time from the quadratic order to linear order with respect to the number of points. In the cosh-attention, the traditional softmax operator is replaced by non-negative ReLU activation and hyperbolic-cosine-based operator with re-weighting mechanism. Then based on the key component, cosh-attention, we present a two-stage hyperbolic cosine transformer (ChTR3D) for 3D object detection from point clouds. It refines proposals by applying cosh-attention in linear computation complexity to encode rich contextual relationships among points. Extensive experiments on the widely used KITTI dataset and Waymo Open Dataset demonstrate that compared with vanilla attention, the cosh-attention significantly improves the inference speed with competitive performance. Among two-stage state-of-the-art methods using point-level features for refinement, the proposed ChTR3D is the fastest one.

Keywords

A

References

[1] Choy C, Gwak J Y, Savarese S. 4D spatio-temporal convnets: Minkowski convolutional neural networks. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 16–20, 2019, Long Beach, CA, USA. 2019, New York, IEEE30753084[C]

[2] Yan Y, Mao Y, Li B. SECOND: sparsely embedded convolutional detection. Sensors. 2018, 18(10): 3337. J]

[3] Deng J, Shi S, Li Pet al. . Voxel R-CNN towards high performance voxel-based 3D object detection. Proceedings of the AAAI Conference on Artificial Intelligence, February 2–9, 2021, Vancouver, Canada. 20211201-1209[C]

[4] Wang J, Lan S, Gao Met al. . Infofocus: 3D object detection for autonomous driving with dynamic information modeling. Proceedings of the European Conference on Computer Vision, August 23–28, 2020, Glasgow, UK. 2020, Heidelberg, Springer405420[C]

[5] Sheng H, Cai S, Liu Yet al. . Improving 3D object detection with channel-wise transformer. Proceedings of the IEEE/CVF International Conference on Computer Vision, October 10–17, 2021, Montreal, Canada. 2021, New York, IEEE27432752[C]

[6] Qi C R, Su H, Mo Ket al. . Pointnet: deep learning on point sets for 3D classification and segmentation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, July 21–26, 2017, Honolulu, HI, USA. 2017, New York, IEEE652660[C]

[7] QI C R, YI L, SU H, et al. Pointnet++: deep hierarchical feature learning on point sets in a metric space[J]. Advances in neural information processing systems, 2017.

[8] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30.

[9] Zhao H, Jiang L, Jia Jet al. . Point transformer. Proceedings of the IEEE/CVF International Conference on Computer Vision, October 10–17, 2021, Montreal, Canada. 2021, New York, IEEE1625916268[C]

[10] Guo M H, Cai J X, Liu Z Net al. . PCT: point cloud transformer. Computational visual media. 2021, 7(2): 187-199. J]

[11] ZHEN Q, SUN W, DENG H, et al. Cosformer: rethinking softmax in attention[EB/OL]. (2022-02-17) [2025-04-26]. https://arxiv.org/abs/2202.08791.

[12] Thomas H, Qi C R, Deschaud J Eet al. . Kpconv: flexible and deformable convolution for point clouds. Proceedings of the IEEE/CVF International Conference on Computer Vision, October 27–November 2, 2019, Seoul, Korea. 2019, New York, IEEE64116420[C]

[13] Shi S, Wang X, Li H. PointrCNN: 3D object proposal generation and detection from point cloud. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 16–20, 2019, Long Beach, CA, USA. 2019, New York, IEEE770779[C]

[14] Yang Z, Sun Y, Liu Set al. . 3DSSD: point-based 3D single stage object detector. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 13–19, 2020, Seattle, WA, USA. 2020, New York, IEEE1104011048[C]

[15] Zhou Y, Tuzel O. Voxelnet: end-to-end learning for point cloud based 3D object detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June 18–23, 2018, Salt Lake City, USA. 2018, New York, IEEE44904499[C]

[16] Shi S, Guo C, Jiang Let al. . PV-RCNN: point-voxel feature set abstraction for 3D object detection. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 13–19, 2020, Seattle, WA, USA. 2020, New York, IEEE1052910538[C]

[17] Hatamizadeh A, Kautz J. MambaVision: a hybrid mamba-transformer vision backbone. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 11–15, 2025, Nashville, TN, USA. 2025, New York, IEEE2526125270[C]

[18] Mao J, Xue Y, Nu Met al. . Voxel transformer for 3D object detection. Proceedings of the IEEE/CVF International Conference on Computer Vision, October 10–17, 2021, Montreal, Canada. 2021, New York, IEEE31643173[C]

[19] Guan T, Wang J, Lan Set al. . M3DETR: multi-representation, multi-scale, mutual-relation 3D object detection with transformers. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, January 3–8, 2022, Waikoloa, HI, USA. 2022, New York, IEEE22932303. C]

[20] OpenPCDet Development Team. OpenPCDet: an opensource toolbox for 3D object detection from point clouds. 2020

[21] Mao J, Nu M, Bai Het al. . Pyramid R-CNN: towards better performance and adaptability for 3D object detection. Proceedings of the IEEE/CVF International Conference on Computer Vision, October 10–17, 2021, Montreal, Canada. 2021, New York, IEEE27232732[C]

[22] Li Z, Yao Y, Quan Zet al. . Spatial information enhancement network for 3D object detection from point cloud. Pattern recognition. 2022, 128: 108684. J]

[23] Hu J S K, Kuai T, Waslander S L. Point density-aware voxels for lidar 3D object detection. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 19–24, 2022, New Orleans, Louisiana, USA. 2022, New York, IEEE8469-8478[C]

RIGHTS & PERMISSIONS

Tianjin University of Technology