Point-to-Voxel Knowledge Distillation for LiDAR Semantic Segmentation

Hou, Yuenan; Zhu, Xinge; Ma, Yuexin; Loy, Chen Change; Li, Yikang

doi:10.1109/cvpr52688.2022.00829

Cited by 100 publications

(48 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These methods become particularly efficient on autonomous driving scenes when, e.g., adapting the shape of the voxels to the point sampling structure [42], and can reach even higher performance when using an encoder with an attentive feature module that extract information about the global context and the local details [6]. Finally, [11] shows that these architectures can be compressed while preserving their performance by using a point-to-voxel knowledge distillation loss.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation

Puy¹,

Boulch²,

Marlet³

2023

Preprint

View full text Add to dashboard Cite

Semantic segmentation of point clouds in autonomous driving datasets requires techniques that can process large numbers of points over large field of views. Today, most deep networks designed for this task exploit 3D sparse convolutions to reduce memory and computational loads. The best methods then further exploit specificities of rotating lidar sampling patterns to further improve the performance, e.g., cylindrical voxels, or range images (for feature fusion from multiple point cloud representations). In contrast, we show that one can build a well-performing point-based backbone free of these specialized tools. This backbone, WaffleIron, relies heavily on generic MLPs and dense 2D convolutions, making it easy to implement, and contains just a few parameters easy to tune. Despite its simplicity, our experiments on SemanticKITTI and nuScenes show that Waf-fleIron competes with the best methods designed specifically for these autonomous driving datasets. Hence, WaffleIron is a strong, easy-to-implement, baseline for semantic segmentation of sparse outdoor point clouds. The code is available at https://github.com/valeoai/WaffleIron.

show abstract

Section: Related Workmentioning

confidence: 99%

“…SemanticKITTI. On this dataset, we adopt training and inference practices used in, e.g., [11,36,37,42]. In particular, the model is trained using both the training split and the validation split and evaluated on the test set.…”

Section: Performance On Autonomous Driving Datasetsmentioning

confidence: 99%

Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation

Puy¹,

Boulch²,

Marlet³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Previous works typically resort to different paradigms for indoor [31,41,43,29,13] and outdoor [35,36,20,33,19,1,49,10] scenes, given their difference in the point cloud distribution. Our work focuses on the outdoor case, where the point clouds are obtained by LiDAR and exhibit sparse and non-uniform distribution.…”

Section: Related Workmentioning

confidence: 99%

Position-Guided Point Cloud Panoptic Segmentation Transformer

Xiao¹,

Zhang²,

Wang³

et al. 2023

Preprint

View full text Add to dashboard Cite

DEtection TRansformer (DETR) started a trend that uses a group of learnable queries for unified visual perception. This work begins by applying this appealing paradigm to LiDAR-based point cloud segmentation and obtains a simple yet effective baseline. Although the naive adaptation obtains fair results, the instance segmentation performance is noticeably inferior to previous works. By diving into the details, we observe that instances in the sparse point clouds are relatively small to the whole scene and often have similar geometry but lack distinctive appearance for segmentation, which are rare in the image domain. Considering instances in 3D are more featured by their positional information, we emphasize their roles during the modeling and design a robust Mixed-parameterized Positional Embedding (MPE) to guide the segmentation process. It is embedded into backbone features and later guides the mask prediction and query update processes iteratively, leading to Position-Aware Segmentation (PA-Seg) and Masked Focal Attention (MFA). All these designs impel the queries to attend to specific regions and identify various instances. The method, named Position-guided Point cloud Panoptic segmentation transFormer (P3Former), outperforms previous state-ofthe-art methods by 3.4% and 1.2% PQ on SemanticKITTI and nuScenes benchmark, respectively. The source code and models are available at https://github.com/SmartBot-PJLab/P3Former.

show abstract

“…Knowledge distillation for point cloud semantic segmentation have proven difficult due to the sparsity, randomness, and varying densiy of point clouds. To this end Point-to-voxel knowledge distillation 40 (PVKD) was proposed. PVKD distills elements from both the point-level and voxel-level, this creates a specific distillation process for point clouds.…”

Section: Fusion and Pruningmentioning

confidence: 99%

Semantic segmentation of point clouds from scanning lidars

Axelsson

Holmberg²,

Tulldahl³

2022

Electro-Optical Remote Sensing XVI

View full text Add to dashboard Cite

A point cloud can provide a detailed three dimensional (3D) description of a scene. Partitioning of a point cloud into semantic classes is important for scene understanding, which can be used in autonomous navigation for unmanned vehicles and in applications including surveillance, mapping, and reconnaissance. In this paper, we give a review of recent machine learning techniques for semantic segmentation of point clouds from scanning lidars and an overview of model compression techniques. We focus especially on scan-based learning approaches, which operate on single sensor sweeps. These methods do not require data registration and are suitable for real-time applications. We demonstrate how these semantic segmentation techniques can be used in defence applications in surveillance or mapping scenarios with a scanning lidar mounted on a small UAV.

show abstract

Point-to-Voxel Knowledge Distillation for LiDAR Semantic Segmentation

Cited by 100 publications

References 22 publications

Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation

Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation

Position-Guided Point Cloud Panoptic Segmentation Transformer

Semantic segmentation of point clouds from scanning lidars

Contact Info

Product

Resources

About