# voxelGenerator Plugin
**Table Of Contents**
- [Description](#description)
* [Structure](#structure)
- [Parameters](#parameters)
- [Additional resources](#additional-resources)
- [License](#license)
- [Changelog](#changelog)
- [Known issues](#known-issues)
## Description
The `voxelGeneratorPlugin` performs the generation of voxels(pillars) from raw points in a point cloud frame. This operation essentially quantize the 3D points in spacial dimensions(x, y, z) with a certain granularity. The output of this plugin will be a group of pillars.
`voxelGeneratorPlugin` implements a quantization of 3D points in point cloud data and produces a groups of voxels. Each voxel is either empty or contains several points that are close to each other.
This plugin is optimized for the above steps and it allows you to do PointPillars inference in TensorRT.
### Structure
The `voxelGeneratorPlugin` takes 2 inputs; `points`, and `num_points`.
`points`
The input raw points from a point cloud. The shape of this tensor is `[N, M, C]`, where `N` is batch size, `M` is the maximum number of points in a point cloud frame, and `C` is the number of channels for each point.
Since point cloud data is sparse in nature, each frame will generally have different number of valid points(no more than `M`). Zero-padding should be applied properly to construct a dense tensor from a batch of point cloud frames.
`num_points`
The number of valid points in each frame. The valid number of points should be no more than `M`. The shape of this tensor is `[N]`.
The `voxelGeneratorPlugin` generates the following 3 outputs:
`voxels`
The voxels generated by this plugin. The shape of this tensor is `[N, V, P, C']`, where `N` is batch size, `V` is the maximum number of voxels(pillars) per frame, `P` is the maximum number of points per voxel, and `C'` is the number of channels(features) per point in voxels.
`voxel_coords`
The coordinates of each voxel in `voxels`. This coordinates tensor will be used to compute a dense feature map indirectly from the `voxels`(after some reduction operations are applied to `voxels`). The shape of this tensor is `[N, V, 4]`, where `N, V` are as above and 4 is just the length of coordinates encoded as `(frame_id, z, y, x)`.
`num_pillar`
The number of valid voxels(pillars) in `voxels` for each frame. This will be used to generate the dense feature map. The shape of this tensor is `[N]`.
## Parameters
`voxelGeneratorPlugin` has plugin creator class `voxelGeneratorPluginCreator` and plugin class `voxelGeneratorPlugin`.
The parameters are defined below and consists of the following attributes:
| Type | Parameter | Description
|----------|--------------------------|--------------------------------------------------------
| `int` | `max_num_points_per_voxel` | Maximum number of points per voxel.
| `int` | `max_voxels` | Maximum number of voxels to be generated per frame.
| `list of floats` | `point_cloud_range` | The range of the point cloud coordinates.
| `int` | `voxel_feature_num` | The number of channels of the generated voxels.
| `list of floats` | `voxel_size` | The size of the voxels.
## Additional resources
The following resources provide a deeper understanding of the `voxelGeneratorPlugin` plugin:
**Networks:**
- [PointPillars](https://arxiv.org/pdf/1812.05784)
**Documentation:**
- [PointPillars](https://arxiv.org/pdf/1812.05784)
没有合适的资源?快使用搜索试试~ 我知道了~
资源推荐
资源详情
资源评论
收起资源包目录
TensorRT-Plugin实现之VoxelGenerator算子实现_cuda_TRT8.zip (57个子文件)
TensorRT-Plugin实现之VoxelGenerator算子实现_cuda_TRT8
common
bboxUtils.h 2KB
serialize.hpp 4KB
templates.h 2KB
CMakeLists.txt 969B
cudaDriverWrapper.h 4KB
plugin.cpp 1KB
vfcCommon.h 1KB
dimsHelpers.h 2KB
mrcnn_config.h 5KB
nmsUtils.h 1KB
cudaDriverWrapper.cpp 5KB
checkMacrosPlugin.cpp 5KB
nmsHelper.cpp 2KB
plugin.h 11KB
common.cuh 13KB
kernels
proposalKernel.cu 25KB
extractFgScores.cu 3KB
CMakeLists.txt 943B
maskRCNNKernels.cu 116KB
saturate.h 1KB
pillarScatterKernels.cu 4KB
kernel.h 15KB
voxelGeneratorKernels.cu 17KB
roiPooling.cu 14KB
detectionForward.cu 8KB
regionForward.cu 5KB
sortScoresPerImage.cu 5KB
decodeBbox3DKernels.cu 7KB
maskRCNNKernels.h 10KB
proposalsForward.cu 6KB
normalizeLayer.cu 10KB
reducedMathPlugin.h 3KB
sortScoresPerClass.cu 11KB
decodeBBoxes.cu 15KB
bboxDeltas2Proposals.cu 11KB
gatherTopDetections.cu 8KB
gridAnchorLayer.cu 5KB
rproiInferenceFused.cu 4KB
generateAnchors.cu 4KB
kernel.cpp 2KB
reorgForward.cu 4KB
allClassNMS.cu 15KB
nmsLayer.cu 16KB
priorBoxLayer.cu 8KB
lReLU.cu 2KB
cropAndResizeKernel.cu 5KB
common.cu 5KB
permuteData.cu 4KB
checkMacrosPlugin.h 15KB
bertCommon.h 17KB
cub_helper.h 1KB
reducedMathPlugin.cpp 3KB
vfcCommon.cpp 2KB
half.h 1KB
voxelGeneratorPlugin
voxelGenerator.cpp 17KB
voxelGenerator.h 4KB
README.md 3KB
共 57 条
- 1
资源评论
极智视界
- 粉丝: 2w+
- 资源: 1600
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- AutoHotKey 2.0中文帮助文件
- 基于Docker-compose的Elasticsearch集群每个节点均是独立docker-compose配置而成源码.zip
- 目标检测-零售食品LOGO检测数据集-40000张图-+对应VOC-COCO-YOLO三种格式标签+数据集划分脚本
- 目标检测-零售食品LOGO检测数据集-30000张图-+对应VOC-COCO-YOLO三种格式标签+数据集划分脚本
- 目标检测-零售食品LOGO检测数据集-20000张图-+对应VOC-COCO-YOLO三种格式标签+数据集划分脚本
- 目标检测-零售食品LOGO检测数据集-10000张图-+对应VOC-COCO-YOLO三种格式标签+数据集划分脚本
- 基于GUI+MYSQL+JAVA图书管理系统文档说明+源码(高分大作业项目).zip
- 基于Qt使用C++实现图书管理系统源码+数据库(95分以上).zip
- 基于GUI+MYSQL+JAVA票务管理系统文档介绍+源码+数据库(高分大作业).zip
- Java项目-购物网站系统(java+Servlet+JSP+Mysql)
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功