A rewrite verson of Lidar detection deeplearning framework (PointPillars) for autonomous-driving (pc or vehicle computer) applications.
[This repo is not maintained， and the overall redundancy is not suitable for deployment on the vehicle].
you can use it repository to achieve fast Lidar detection in your autoware device. (only test in Nvidia Xavier: each frame process less than 50 ms!)
Pointpillars demonstrates how to reproduce the results from PointPillars: Fast Encoders for Object Detection from Point Clouds (to be published at CVPR 2019) on the KITTI dataset by making the minimum required changes from the preexisting open source codebase SECOND.
This is not an official nuTonomy codebase, but it can be used to match the published PointPillars results.
WARNING: This code is not being actively maintained. This code can be used to reproduce the results in the first version of the paper, https://arxiv.org/abs/1812.05784v1. For an actively maintained repository that can also reproduce PointPillars results on nuScenes, we recommend using SECOND. We are not the owners of the repository, but we have worked with the author and endorse his code.
1: Training and evaluating on your GPU device with Pytorch to get the suitable weights
||
||
\/
2: Transfer the original submodels (with weights)to tensorrt version(pfn.trt and bankbone.trt).
||
||
\/
3: Detecting objects of original pointcloud (x,y,z,intensity) on vehicle device.
1: Training and evaluating on your GPU device with Pytorch to get the suitable weights
||
\/
2: Conversion of Weights to a model checkpoint { pfn.trt and bankbone.trt }
||
\/
3: Create TensorRT Network
||
\/
4: Create (optional: additional) layers for analysis
||
\/
5: Build the tensorRT Inference Engine (Quantization, Decoding, AMT-FP16 conversion etc),
||
\/
6: Retrieving the engine from inference Engine
||
\/
7: Deploy Model
├── core
├── data
├── docs
├── libs
│ ├── ops
│ │ ├── cc
│ │ │ └── nms
│ │ ├── non_max_suppression
│ │ └── point_cloud
│ └── tools
│ └── buildtools
├── logs
├── models
│ ├── bones <------ The sub-modules list here
│ └── detectors <------ The Main network lies here
└── params
├── configs
├── {./Path/to/your TensorRT files(.trt)}
└── {./Path/to/your weights files(.ckpt)}
Device 1: NVIDIA GeForce 2070Ti：
├── SM-75
└── 4GB or more of memory
Device 2: NVIDIA Jstson AGX xavier:
└── SM-72
- ONLY supports python 3.6+, pytorch 1.1+, Ubuntu 18.04.
- CUDA 9.0+
- CuDNN 7+
- TensorRT 6.0 (only need for xavier)
1.Refer to TRAIN.md for the installation of the training stage of PointPillars on 2070Ti.
2.Refer to INFERENCE.md for the installation of the inference stage of PointPillars on xavier.
#Performance
This is mainly due to TensorRT, which makes the network runs four times faster than the original version.(test on Nvidia Xavier)
Emmmm.....It seem doesn't look bad either..lmao.
