一、官方原代码
#!/usr/bin/env python3
#
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
# DEALINGS IN THE SOFTWARE.
#
import sys
import argparse
from jetson_inference import segNet
from jetson_utils import videoSource, videoOutput, cudaOverlay, cudaDeviceSynchronize, Log
from segnet_utils import *
# parse the command line
parser = argparse.ArgumentParser(description="Segment a live camera stream using an semantic segmentation DNN.",
formatter_class=argparse.RawTextHelpFormatter,
epilog=segNet.Usage() + videoSource.Usage() + videoOutput.Usage() + Log.Usage())
parser.add_argument("input", type=str, default="", nargs='?', help="URI of the input stream")
parser.add_argument("output", type=str, default="", nargs='?', help="URI of the output stream")
parser.add_argument("--network", type=str, default="fcn-resnet18-voc", help="pre-trained model to load, see below for options")
parser.add_argument("--filter-mode", type=str, default="linear", choices=["point", "linear"], help="filtering mode used during visualization, options are:\n 'point' or 'linear' (default: 'linear')")
parser.add_argument("--visualize", type=str, default="overlay,mask", help="Visualization options (can be 'overlay' 'mask' 'overlay,mask'")
parser.add_argument("--ignore-class", type=str, default="void", help="optional name of class to ignore in the visualization results (default: 'void')")
parser.add_argument("--alpha", type=float, default=150.0, help="alpha blending value to use during overlay, between 0.0 and 255.0 (default: 150.0)")
parser.add_argument("--stats", action="store_true", help="compute statistics about segmentation mask class output")
try:
args = parser.parse_known_args()[0]
except:
print("")
parser.print_help()
sys.exit(0)
# load the segmentation network
net = segNet(args.network, sys.argv)
# note: to hard-code the paths to load a model, the following API can be used:
#
# net = segNet(model="model/fcn_resnet18.onnx", labels="model/labels.txt", colors="model/colors.txt",
# input_blob="input_0", output_blob="output_0")
# set the alpha blending value
net.SetOverlayAlpha(args.alpha)
# create video output
output = videoOutput(args.output, argv=sys.argv)
# create buffer manager
buffers = segmentationBuffers(net, args)
# create video source
input = videoSource(args.input, argv=sys.argv)
# process frames until EOS or the user exits
while True:
# capture the next image
img_input = input.Capture()
if img_input is None: # timeout
continue
# allocate buffers for this size image
buffers.Alloc(img_input.shape, img_input.format)
# process the segmentation network
net.Process(img_input, ignore_class=args.ignore_class)
# generate the overlay
if buffers.overlay:
net.Overlay(buffers.overlay, filter_mode=args.filter_mode)
# generate the mask
if buffers.mask:
net.Mask(buffers.mask, filter_mode=args.filter_mode)
# composite the images
if buffers.composite:
cudaOverlay(buffers.overlay, buffers.composite, 0, 0)
cudaOverlay(buffers.mask, buffers.composite, buffers.overlay.width, 0)
# render the output image
output.Render(buffers.output)
# update the title bar
output.SetStatus("{:s} | Network {:.0f} FPS".format(args.network, net.GetNetworkFPS()))
# print out performance info
cudaDeviceSynchronize()
net.PrintProfilerTimes()
# compute segmentation class stats
if args.stats:
buffers.ComputeStats()
# exit on input/output EOS
if not input.IsStreaming() or not output.IsStreaming():
break
二、代码分析
这段Python代码是用于使用语义分割深度神经网络(DNN)对实时摄像头流进行分割。代码主要基于NVIDIA的Jetson平台。下面是对代码的详细分析:
代码概述
1. 头部和导入模块
#!/usr/bin/env python3
#
# 版权声明
#
# 导入必要的模块
import sys
import argparse
from jetson_inference import segNet
from jetson_utils import videoSource, videoOutput, cudaOverlay, cudaDeviceSynchronize, Log
from segnet_utils import *
- 头部信息:包含shebang行,用于指定使用Python 3运行脚本。
- 版权声明:声明了NVIDIA公司的版权信息,并详细说明了使用、修改、分发软件的许可条件。
- 导入模块:导入了系统模块和Jetson平台的相关工具模块。
2. 命令行参数解析
# 解析命令行参数
parser = argparse.ArgumentParser(description="使用语义分割DNN对实时摄像头流进行分割。",
formatter_class=argparse.RawTextHelpFormatter,
epilog=segNet.Usage() + videoSource.Usage() + videoOutput.Usage() + Log.Usage())
parser.add_argument("input", type=str, default="", nargs='?', help="输入流的URI")
parser.add_argument("output", type=str, default="", nargs='?', help="输出流的URI")
parser.add_argument("--network", type=str, default="fcn-resnet18-voc", help="预训练模型的名称")
parser.add_argument("--filter-mode", type=str, default="linear", choices=["point", "linear"], help="可视化过程中使用的过滤模式")
parser.add_argument("--visualize", type=str, default="overlay,mask", help="可视化选项(可以是'overlay', 'mask', 'overlay,mask')")
parser.add_argument("--ignore-class", type=str, default="void", help="在可视化结果中忽略的类别名称(默认:'void')")
parser.add_argument("--alpha", type=float, default=150.0, help="叠加时使用的alpha混合值")
parser.add_argument("--stats", action="store_true", help="计算分割掩码类别输出的统计信息")
try:
args = parser.parse_known_args()[0]
except:
print("")
parser.print_help()
sys.exit(0)
- 参数解析:使用argparse模块解析命令行参数,用户可以通过命令行指定输入、输出流、模型名称、过滤模式等。
3. 加载分割网络
# 加载分割网络
net = segNet(args.network, sys.argv)
- 加载模型:根据解析的参数加载指定的预训练模型。
4. 设置混合值
# 设置叠加的alpha值
net.SetOverlayAlpha(args.alpha)
- 设置alpha混合值:用于在可视化时控制叠加效果的透明度。
5
5. 创建视频输入和输出对象
# 创建视频输出
output = videoOutput(args.output, argv=sys.argv)
# 创建缓冲区管理器
buffers = segmentationBuffers(net, args)
# 创建视频源
input = videoSource(args.input, argv=sys.argv)
- 视频输出:创建用于显示或保存输出视频的对象。
- 缓冲区管理器:管理用于存储和处理图像数据的缓冲区。
- 视频输入:创建用于捕获输入视频流的对象。
6. 处理视频帧
# 处理视频帧直到结束或用户退出
while True:
# 捕获下一帧图像
img_input = input.Capture()
if img_input is None: # 超时处理
continue
# 为当前尺寸的图像分配缓冲区
buffers.Alloc(img_input.shape, img_input.format)
# 处理分割网络
net.Process(img_input, ignore_class=args.ignore_class)
# 生成叠加图像
if buffers.overlay:
net.Overlay(buffers.overlay, filter_mode=args.filter_mode)
# 生成掩码图像
if buffers.mask:
net.Mask(buffers.mask, filter_mode=args.filter_mode)
# 合成图像
if buffers.composite:
cudaOverlay(buffers.overlay, buffers.composite, 0, 0)
cudaOverlay(buffers.mask, buffers.composite, buffers.overlay.width, 0)
# 渲染输出图像
output.Render(buffers.output)
# 更新标题栏状态
output.SetStatus("{:s} | Network {:.0f} FPS".format(args.network, net.GetNetworkFPS()))
# 输出性能信息
cudaDeviceSynchronize()
net.PrintProfilerTimes()
# 计算分割类别统计信息
if args.stats:
buffers.ComputeStats()
# 检查输入/输出流是否在流动
if not input.IsStreaming() or not output.IsStreaming():
break
- 捕获图像:从输入视频流中捕获一帧图像。
- 缓冲区分配:为当前图像的尺寸和格式分配缓冲区。
- 处理分割网络:对捕获的图像进行语义分割处理。
- 生成叠加和掩码图像:根据处理结果生成叠加和掩码图像。
- 图像合成:将叠加和掩码图像合成为最终输出图像。
- 渲染输出图像:将合成的图像显示或保存。
- 更新状态和性能信息:更新输出窗口的状态栏,显示当前网络的FPS(每秒帧数),并输出性能信息。
- 计算统计信息:如果启用了统计选项,则计算分割掩码的类别统计信息。
- 检查流状态:检查输入和输出流是否仍在流动,如果任一流停止,则退出循环。
结论
这段代码的主要功能是通过NVIDIA Jetson平台上的语义分割深度神经网络处理实时摄像头流,并生成分割后的可视化结果。代码包含了从参数解析、模型加载、图像捕获处理到最终输出的完整流程,适合用于实时视频流的语义分割任务。