cpp目录下提供了C++例程以供参考使用,具体情况如下:
| 序号 | C++例程 | 说明 |
|---|---|---|
| 1 | superglue_bmcv | 使用OpenCV解码、BMCV前处理、BMRT推理 |
如果您在x86/arm平台安装了PCIe加速卡(如SC系列加速卡),可以直接使用它作为开发环境和运行环境。您需要安装libsophon、sophon-opencv和sophon-ffmpeg,具体步骤可参考x86-pcie平台的开发和运行环境搭建。
如果您使用SoC平台(如SE、SM系列边缘设备),刷机后在/opt/sophon/下已经预装了相应的libsophon、sophon-opencv和sophon-ffmpeg运行库包,可直接使用它作为运行环境。通常还需要一台x86主机作为开发环境,用于交叉编译C++程序。
C++程序运行前需要编译可执行文件。
安装以下第三方依赖:
sudo apt install libopenblas-dev可以直接在PCIe平台上编译程序:
cd cpp/superglue_bmcv
mkdir build && cd build
cmake ..
make
cd ..编译完成后,会在superglue_bmcv目录下生成superglue_bmcv.pcie。
通常在x86主机上(本例程使用ubuntu20.04/22.04系统)交叉编译程序,您需要在x86主机上使用SOPHON SDK搭建交叉编译环境,将程序所依赖的头文件和库文件打包至soc-sdk目录中,具体请参考交叉编译环境搭建。本例程主要依赖libsophon、sophon-opencv和sophon-ffmpeg运行库包以及一些第三方库,为了在x86上下载这些arm64架构的第三方库,您可能需要更换apt源,将如下源替换掉您/etc/apt/source.list里的源,注意对原来的源做好备份。
#for ubuntu 20.04
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ focal main restricted universe multiverse
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ focal-security main restricted universe multiverse
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ focal-updates main restricted universe multiverse
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
#for ubuntu 22.04
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ jammy main restricted universe multiverse
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ jammy-security main restricted universe multiverse
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ jammy-updates main restricted universe multiverse
deb [arch=amd64] https://mirrors.aliyun.com/ubuntu/ jammy-backports main restricted universe multiverse
然后在/etc/apt/sources.list.d/目录下新建arm-cross-compile-sources.list文件,写入如下内容:
#for ubuntu 20.04
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ focal main restricted universe multiverse
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ focal-security main restricted universe multiverse
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ focal-updates main restricted universe multiverse
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ focal-backports main restricted universe multiverse
#for ubuntu 22.04
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ jammy main restricted universe multiverse
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ jammy-security main restricted universe multiverse
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ jammy-updates main restricted universe multiverse
deb [arch=arm64] https://mirrors.aliyun.com/ubuntu-ports/ jammy-backports main restricted universe multiverse然后运行如下命令,下载第三方依赖:
#如果当前环境在stream_dev的docker镜像中,可以不用加sudo
sudo dpkg --add-architecture arm64
sudo apt update
sudo apt install libopenblas-dev:arm64
sudo apt install ccache:arm64
sudo apt install numactl:arm64
sudo apt install libhwloc-dev:arm64
sudo apt install libevent-dev:arm64交叉编译环境搭建好后,使用交叉编译工具链编译生成可执行文件:
cd cpp/superglue_bmcv
mkdir build && cd build
#请根据实际情况修改-DSDK的路径,需使用绝对路径。
cmake -DTARGET_ARCH=soc -DSDK=/path_to_sdk/soc-sdk ..
make编译完成后,会在superglue_bmcv目录下生成superglue_bmcv.soc。
对于PCIe平台,可以直接在PCIe平台上推理测试;对于SoC平台,需将交叉编译生成的可执行文件及所需的模型、测试数据拷贝到SoC平台中测试。测试的参数及运行方式是一致的,下面主要以PCIe模式进行介绍。
注意,如果是在SoC平台上进行测试,需要在SoC上安装如下第三方依赖并设置环境变量:
sudo apt install libopenblas-dev
export LD_LIBRARY_PATH=$PWD/../aarch64_lib/libtorch/lib:$LD_LIBRARY_PATH可执行程序默认有一套参数,请注意根据实际情况进行传参,superglue_bmcv.pcie为例,具体参数说明如下:
Usage: superglue_bmcv.pcie [params]
--bmodel_superglue (value:../../models/BM1688/superglue_fp32_1b_1024.bmodel)
Path to bmodel.
--bmodel_superpoint (value:../../models/BM1688/superpoint_fp32_1b.bmodel)
Path to bmodel.
--dev_id (value:0)
TPU device id.
--help (value:true)
Print help information.
--input_dir (value:../../datasets/scannet_sample_images)
Path to the directory that contains the images.
--input_pairs (value:../../datasets/scannet_sample_pairs_with_gt.txt)
Path to the list of image pairs.
--keypoint_thresh (value:0.0002)
Keypoint threshold.
--matching_thresh (value:0.002)
Matching threshold.
--max_keypoint_size (value:1024)
Max keypoint size.
--nms_radius (value:4)
Nms radius.图片测试实例如下:
./superglue_bmcv.soc --bmodel_superglue=../../models/BM1688/superglue_fp32_1b_iter20_1024.bmodel --bmodel_superpoint=../../models/BM1688/superpoint_fp32_1b.bmodel --input_dir=../../datasets/scannet_sample_images --input_pairs=../../datasets/scannet_sample_pairs_with_gt.txt测试结束后,会将图片保存在results/images下,匹配结果保存在results/result.json下,同时会打印性能信息。
