GstInference/Benchmarks: Difference between revisions

From RidgeRun Developer Wiki
(Created page with "= Benchmarks = The following benchmarks were run with a source video (1920x1080@60). With the following base gstremear pipeline, and environment variables: <source lang="bash...")
 
No edit summary
Tag: Manual revert
 
(187 intermediate revisions by 7 users not shown)
Line 1: Line 1:
= Benchmarks =
<noinclude>
The following benchmarks were run with a source video (1920x1080@60). With the following base gstremear pipeline, and environment variables:
{{GstInference/Head|previous=Example Applications/DispTec|next=Model Zoo|metakeywords=GstInference gstreamer pipelines, Inference gstreamer pipelines, NCSDK Inference gstreamer pipelines, GoogLeNet, TinyYolo v2, GoogLeNet x2, TensorFlow backend|title=GstInference Benchmarks}}
</noinclude>
 
 
<html>
<script
/* This should be added once per page only */
type="text/javascript" src="https://www.gstatic.com/charts/loader.js">
</script>
</html>
 
== GstInference Benchmarks - Introduction ==
 
This wiki summarizes a series of benchmarks on different hardware platforms based on the [https://github.com/RidgeRun/gst-inference/blob/master/tests/benchmark/run_benchmark.sh run_benchmark.sh] bash script that can be found in the official [https://github.com/RidgeRun/gst-inference GstInference repository]. The script is based on the following GStreamer pipeline:


<source lang="bash">
<source lang="bash">
$ VIDEO_FILE='video.mp4'
#Script to run each model
$ MODEL_LOCATION='graph_inceptionv1_tensorflow.pb'
run_all_models(){
$ INPUT_LAYER='input'
 
$ OUTPUT_LAYER='InceptionV1/Logits/Predictions/Reshape_1'
  model_array=(inceptionv1 inceptionv2 inceptionv3 inceptionv4 tinyyolov2 tinyyolov3)
  model_upper_array=(InceptionV1 InceptionV2 InceptionV3 InceptionV4 TinyYoloV2 TinyYoloV3)
  input_array=(input input input input input/Placeholder inputs )
  output_array=(InceptionV1/Logits/Predictions/Reshape_1 Softmax InceptionV3/Predictions/Reshape_1
  InceptionV4/Logits/Predictions add_8 output_boxes )
 
  mkdir -p logs/
  rm -f logs/*
 
  for ((i=0;i<${#model_array[@]};++i)); do
    echo Perf ${model_array[i]}
    gst-launch-1.0 \
    filesrc location=$VIDEO_PATH num-buffers=600 ! decodebin ! videoconvert ! \
    perf print-arm-load=true name=inputperf ! tee name=t t. ! videoscale ! queue ! net.sink_model t. ! queue ! net.sink_bypass \
    ${model_array[i]} backend=$BACKEND name=net backend::input-layer=${input_array[i]} backend::output-layer=${output_array[i]} \
    model-location="${MODELS_PATH}${model_upper_array[i]}_${INTERNAL_PATH}/graph_${model_array[i]}${EXTENSION}" \
    net.src_bypass ! perf print-arm-load=true name=outputperf ! videoconvert ! fakesink sync=false > logs/${model_array[i]}.log
  done
}
</source>
 
=== Test benchmark video ===
The following video was used to perform the benchmark tests.
<br>
To download the video press right-click on the video and select 'Save video as' and save this on your computer.
<br>
<br>
[[File:Test benchmark video.mp4|thumb|border|center|500px|alt=Alt|Video 1. Test benchmark video]]
 
== x86 ==
 
The Desktop PC had the following specifications:
*Intel(R) Core(TM) Core i7-7700HQ CPU @ 2.80GHz
*12 GB RAM
*Linux 4.15.0-106-generic x86_64 (Ubuntu 16.04)
*GStreamer 1.8.3
 
=== FPS Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_fps_x86" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffx86Fps);
     
      function drawStuffx86Fps() {
 
        var chartDiv_Fps_x86 = document.getElementById('chart_fps_x86');
 
        var table_models_fps_x86 = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'ONNXRT \n x86',
          'ONNXRT OpenVINO (CPU_FP32) \n x86',
          'ONNXRT OpenVINO (GPU_FP32)\n x86',
          'ONNXRT OpenVINO (GPU_FP16)\n x86',
          'ONNXRT OpenVINO (MYRIAD_FP16)\n x86',
          'TensorFlow \n x86',
          'TensorFlow Lite \n x86'],
          ['InceptionV1', 47.9, 81.966, 70.580, 98.742, 46.294, 55.3182, 18.8422], //row 1
          ['InceptionV2', 32.7, 63.352, 54.159, 77.449, 34.613, 39.6438, 13.5714], //row 2
          ['InceptionV3', 12.1, 23.287, 20.878, 34.059, 11.999, 16.2488, 4.9924], //row 3
          ['InceptionV4', 5.26, 10.927, 6.160, 4.548, 6.494, 7.793, 2.583], //row 4
          ['TinyYoloV2', 33.559, 32.587, 0, 0, 0, 18.1846, 7.2708], //row 5
          ['TinyYoloV3', 35.092, 27.799, 0, 0, 0, 21.7334, 7.3042]  //row 6
        ]);
        var x86_materialOptions_fps = {
          width: 1000,
          chart: {
            title: 'Model Vs FPS per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_x86_fps = new google.charts.Bar(chartDiv_Fps_x86);
        view_x86_fps = new google.visualization.DataView(table_models_fps_x86);
 
        function drawMaterialChart() {
          var materialChart_x86_fps = new google.charts.Bar(chartDiv_Fps_x86);
          materialChart_x86_fps.draw(table_models_fps_x86, google.charts.Bar.convertOptions(x86_materialOptions_fps));
 
          init_charts();
        }
        function init_charts(){
          view_x86_fps.setColumns([0,1, 2, 3, 4, 5, 6, 7]);
          materialChart_x86_fps.draw(view_x86_fps, x86_materialOptions_fps);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
=== CPU Load Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_cpu_x86" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffx86Cpu);
     
      function drawStuffx86Cpu() {
 
        var chartDiv_Cpu_x86 = document.getElementById('chart_cpu_x86');
 
        var table_models_cpu_x86 = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'ONNXRT \n x86',
          'ONNXRT OpenVINO (CPU_FP32) \n x86',
          'ONNXRT OpenVINO (GPU_FP32)\n x86',
          'ONNXRT OpenVINO (GPU_FP16)\n x86',
          'ONNXRT OpenVINO (MYRIAD_FP16)\n x86',
          'TensorFlow \n x86',
          'TensorFlow Lite \n x86'],        //Column 1
          ['InceptionV1', 94.6, 49, 31, 29, 14, 74.2, 47.6], //row 1
          ['InceptionV2', 100, 52, 28, 29, 11, 74.2, 43.6], //row 2
          ['InceptionV3', 95.2, 49, 28, 28, 13, 81, 60.2], //row 3
          ['InceptionV4', 88.8, 49, 33, 46, 11, 86, 50], //row 4
          ['TinyYoloV2',  94, 50, 0, 0, 0, 80.6, 46], //row 5
          ['TinyYoloV3',  91.4, 46, 0, 0, 0, 74.6, 42.4]  //row 6
        ]);
        var x86_materialOptions_cpu = {
          width: 1000,
          chart: {
            title: 'Model Vs CPU Load per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'CPU Load'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_x86_cpu = new google.charts.Bar(chartDiv_Cpu_x86);
        view_x86_cpu = new google.visualization.DataView(table_models_cpu_x86);
 
        function drawMaterialChart() {
          var materialChart_x86_cpu = new google.charts.Bar(chartDiv_Cpu_x86);
          materialChart_x86_cpu.draw(table_models_cpu_x86, google.charts.Bar.convertOptions(x86_materialOptions_cpu));
 
          init_charts();
        }
        function init_charts(){
          view_x86_cpu.setColumns([0,1, 2, 3, 4, 5, 6, 7]);
          materialChart_x86_cpu.draw(view_x86_cpu, x86_materialOptions_cpu);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
== Jetson AGX Xavier ==
 
The Jetson Xavier power modes used were 2 and 6 (more information: [https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%2520Linux%2520Driver%2520Package%2520Development%2520Guide%2Fpower_management_jetson_xavier.html%23wwpID0E0OM0HA Supported Modes and Power Efficiency])
 
*View current power mode:
 
<source lang="bash">
$ sudo /usr/sbin/nvpmodel -q
</source>
</source>
The environment variables were changed accordingly with the used model (Inception V1,V2,V3 or V4)
 
*Change current power mode:


<source lang="bash">
<source lang="bash">
$ GST_DEBUG=inception1:1 gst-launch-1.0 filesrc location=$VIDEO_FILE ! decodebin ! videoconvert ! videoscale ! queue ! net.sink_model inceptionv1 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER net.src_model ! perf ! fakesink -v
sudo /usr/sbin/nvpmodel -m x
</source>
</source>
Where x is the power mode ID (e.g. 0, 1, 2, 3, 4, 5, 6).
=== FPS Measurements ===


The Desktop PC had the following specifications:
<html>
*Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz
 
*8 GB RAM
<style>
*Cedar [Radeon HD 5000/6000/7350/8350 Series]
    .button {
*Linux 4.15.0-54-generic x86_64 (Ubuntu 16.04)
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_fps_xavier" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffXavierFps);
     
      function drawStuffXavierFps() {
 
        var chartDiv_Fps_Xavier = document.getElementById('chart_fps_xavier');
 
        var table_models_fps_xavier = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorFlow \n Xavier (15 W)',      //Column 1
          'TensorFlow (GPU) \n Xavier (15 W)', //Column 2
          'TensorFlow \n Xavier (30 W)',      //Column 3
          'TensorFlow (GPU) \n Xavier (30 W)', //Column 4
          'TensorRT \n Xavier',              //Column 5
          'ONNXRT ACL \n Xavier'],            //Column 6
          ['InceptionV1', 8.24, 52.3, 6.41, 66.27, 92.6, 17.566], //row 1
          ['InceptionV2', 6.58, 39.6, 5.11, 50.59, 0, 12.729], //row 2
          ['InceptionV3', 2.54, 17.8, 1.96, 22.95, 24.9, 5.709], //row 3
          ['InceptionV4', 1.22, 9.4, 0.98, 12.14, 13.6, 2.747], //row 4
          ['TinyYoloV2',  0, 0, 0, 0, 69.7, 9.367], //row 5
          ['TinyYoloV3',  0, 0, 0, 0, 0, 10.520]  //row 6
        ]);
        var xavier_materialOptions_fps = {
          width: 900,
          chart: {
            title: 'Model Vs FPS per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_xavier_fps = new google.charts.Bar(chartDiv_Fps_Xavier);
        view_xavier_fps = new google.visualization.DataView(table_models_fps_xavier);
 
        function drawMaterialChart() {
          var materialChart_xavier_fps = new google.charts.Bar(chartDiv_Fps_Xavier);
          materialChart_xavier_fps.draw(table_models_fps_xavier, google.charts.Bar.convertOptions(xavier_materialOptions_fps));
 
          init_charts();
        }
        function init_charts(){
          view_xavier_fps.setColumns([0,1,2,3,4,5,6]);
          materialChart_xavier_fps.draw(view_xavier_fps, xavier_materialOptions_fps);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
=== CPU Load Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_cpu_xavier" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffXavierCpu);
     
      function drawStuffXavierCpu() {
 
        var chartDiv_Cpu_Xavier = document.getElementById('chart_cpu_xavier');
 
        var table_models_cpu_xavier = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorFlow \n Xavier (15 W)',      //Column 1
          'TensorFlow (GPU) \n Xavier (15 W)', //Column 2
          'TensorFlow \n Xavier (30 W)',      //Column 3
          'TensorFlow (GPU) \n Xavier (30 W)', //Column 4
          'TensorRT \n Xavier',              //Column 5
          'ONNXRT ACL \n Xavier'],            //Column 6
          ['InceptionV1', 86, 72, 93, 72, 32, 50], //row 1
          ['InceptionV2', 88, 62.6, 95, 62, 0, 49], //row 2
          ['InceptionV3', 92, 44, 98, 44, 6, 50], //row 3
          ['InceptionV4', 94, 32, 99, 32, 3, 50], //row 4
          ['TinyYoloV2',  0, 0, 0, 0, 16, 50], //row 5
          ['TinyYoloV3',  0, 0, 0, 0, 0, 50]  //row 6
        ]);
        var xavier_materialOptions_cpu = {
          width: 900,
          chart: {
            title: 'Model Vs CPU load per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'CPU Load'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_xavier_cpu = new google.charts.Bar(chartDiv_Cpu_Xavier);
        view_xavier_cpu = new google.visualization.DataView(table_models_cpu_xavier);
 
        function drawMaterialChart() {
          var materialChart_xavier_cpu = new google.charts.Bar(chartDiv_Cpu_Xavier);
          materialChart_xavier_cpu.draw(table_models_cpu_xavier, google.charts.Bar.convertOptions(xavier_materialOptions_cpu));
 
          init_charts();
        }
        function init_charts(){
          view_xavier_cpu.setColumns([0,1,2,3,4,5,6]);
          materialChart_xavier_cpu.draw(view_xavier_cpu, xavier_materialOptions_cpu);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
== Jetson TX2 ==
 
=== FPS Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_fps_tx2" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffTx2Fps);
     
      function drawStuffTx2Fps() {
 
        var chartDiv_Fps_Tx2 = document.getElementById('chart_fps_tx2');
 
        var table_models_fps_tx2 = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorRT \n TX2'],        //Column 1
          ['InceptionV1', 62.5], //row 1
          ['InceptionV2', 0], //row 2
          ['InceptionV3', 20.2], //row 3
          ['InceptionV4', 11.22], //row 4
          ['TinyYoloV2',  33.8], //row 5
          ['TinyYoloV3',  0]  //row 6
        ]);
        var Tx2_materialOptions_fps = {
          width: 900,
          chart: {
            title: 'Model Vs FPS per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_tx2_fps = new google.charts.Bar(chartDiv_Fps_Tx2);
        view_tx2_fps = new google.visualization.DataView(table_models_fps_tx2);
 
        function drawMaterialChart() {
          var materialChart_tx2_fps = new google.charts.Bar(chartDiv_Fps_Tx2);
          materialChart_tx2_fps.draw(table_models_fps_tx2, google.charts.Bar.convertOptions(Tx2_materialOptions_fps));
 
          init_charts();
        }
        function init_charts(){
          view_tx2_fps.setColumns([0,1]);
          materialChart_tx2_fps.draw(view_tx2_fps, Tx2_materialOptions_fps);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
=== CPU Load Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_cpu_tx2" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffTx2Cpu);
     
      function drawStuffTx2Cpu() {
 
        var chartDiv_Cpu_Tx2 = document.getElementById('chart_cpu_tx2');
 
        var table_models_cpu_tx2 = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorRT \n TX2'],        //Column 1
          ['InceptionV1', 21], //row 1
          ['InceptionV2', 0], //row 2
          ['InceptionV3', 10], //row 3
          ['InceptionV4', 7], //row 4
          ['TinyYoloV2',  19], //row 5
          ['TinyYoloV3',  0]  //row 6
        ]);
        var Tx2_materialOptions_cpu = {
          width: 900,
          chart: {
            title: 'Model Vs CPU Load per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_tx2_cpu = new google.charts.Bar(chartDiv_Cpu_Tx2);
        view_tx2_cpu = new google.visualization.DataView(table_models_cpu_tx2);
 
        function drawMaterialChart() {
          var materialChart_tx2_cpu = new google.charts.Bar(chartDiv_Cpu_Tx2);
          materialChart_tx2_cpu.draw(table_models_cpu_tx2, google.charts.Bar.convertOptions(Tx2_materialOptions_cpu));
 
          init_charts();
        }
        function init_charts(){
          view_tx2_cpu.setColumns([0,1]);
          materialChart_tx2_cpu.draw(view_tx2_cpu, Tx2_materialOptions_cpu);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
== Jetson Nano ==
 
=== FPS Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_fps_nano" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffNanoFps);
     
      function drawStuffNanoFps() {
 
        var chartDiv_Fps_Nano = document.getElementById('chart_fps_nano');
 
        var table_models_fps_nano = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorRT \n Nano'],        //Column 1
          ['InceptionV1', 36.8], //row 1
          ['InceptionV2', 0], //row 2
          ['InceptionV3', 10.9], //row 3
          ['InceptionV4', 4.82], //row 4
          ['TinyYoloV2',  14.6], //row 5
          ['TinyYoloV3',  0]  //row 6
        ]);
        var Nano_materialOptions_fps = {
          width: 900,
          chart: {
            title: 'Model Vs FPS per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_nano_fps = new google.charts.Bar(chartDiv_Fps_Nano);
        view_nano_fps = new google.visualization.DataView(table_models_fps_nano);
 
        function drawMaterialChart() {
          var materialChart_nano_fps = new google.charts.Bar(chartDiv_Fps_Nano);
          materialChart_nano_fps.draw(table_models_fps_nano, google.charts.Bar.convertOptions(Nano_materialOptions_fps));
 
          init_charts();
        }
        function init_charts(){
          view_nano_fps.setColumns([0,1]);
          materialChart_nano_fps.draw(view_nano_fps, Nano_materialOptions_fps);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
=== CPU Load Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_cpu_nano" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffNanoCpu);
     
      function drawStuffNanoCpu() {
 
        var chartDiv_Cpu_Nano = document.getElementById('chart_cpu_nano');
 
        var table_models_cpu_nano = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorRT \n Nano'],        //Column 1
          ['InceptionV1', 18], //row 1
          ['InceptionV2', 0], //row 2
          ['InceptionV3', 18], //row 3
          ['InceptionV4', 20], //row 4
          ['TinyYoloV2',  20], //row 5
          ['TinyYoloV3',  0]  //row 6
        ]);
        var Nano_materialOptions_cpu = {
          width: 900,
          chart: {
            title: 'Model Vs CPU Load per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_nano_cpu = new google.charts.Bar(chartDiv_Cpu_Nano);
        view_nano_cpu = new google.visualization.DataView(table_models_cpu_nano);
 
        function drawMaterialChart() {
          var materialChart_nano_cpu = new google.charts.Bar(chartDiv_Cpu_Nano);
          materialChart_nano_cpu.draw(table_models_cpu_nano, google.charts.Bar.convertOptions(Nano_materialOptions_cpu));
 
          init_charts();
        }
        function init_charts(){
          view_nano_cpu.setColumns([0,1]);
          materialChart_nano_cpu.draw(view_nano_cpu, Nano_materialOptions_cpu);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
== Google Coral ==
 
The following benchmarks were performed on the Coral Dev Board.
 
=== FPS Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_fps_coral" style="margin: auto; width: 800px; height: 500px;"></div>
 
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffCoralFps);
     
      function drawStuffCoralFps() {
 
        var chartDiv_Fps_Coral = document.getElementById('chart_fps_coral');
 
        var table_models_fps_coral = google.visualization.arrayToDataTable([
          ['Model',                      //Column 0
          'TensorFlow Lite \n Coral',
          'TensorFlow Lite EdgeTPU \n Coral'],        //Column 1
          ['InceptionV1', 3.11, 41.6], //row 1
          ['InceptionV2', 2.31, 42.8], //row 2
          ['InceptionV3', 0.9, 15.02], //row 3
          ['InceptionV4', 0, 8.56], //row 4
          ['MobileNetV2',  0, 41.12], //row 5
          ['MobileNetV2 + SSD',  0, 38.64]  //row 6
        ]);
        var Coral_materialOptions_fps = {
          width: 900,
          chart: {
            title: 'Model Vs FPS per backend',
          },
          series: {
          },
          axes: {
            y: {
              distance: {side: 'left',label: 'FPS'}, // Left y-axis.
            }
          }
        };
 
        var materialChart_coral_fps = new google.charts.Bar(chartDiv_Fps_Coral);
        view_coral_fps = new google.visualization.DataView(table_models_fps_coral);
 
        function drawMaterialChart() {
          var materialChart_coral_fps = new google.charts.Bar(chartDiv_Fps_Coral);
          materialChart_coral_fps.draw(table_models_fps_coral, google.charts.Bar.convertOptions(Coral_materialOptions_fps));
 
          init_charts();
        }
        function init_charts(){
          view_coral_fps.setColumns([0,1, 2]);
          materialChart_coral_fps.draw(view_coral_fps, Coral_materialOptions_fps);
        }
        drawMaterialChart();
        }
 
</script>
 
</html>
 
=== CPU Load Measurements ===
 
<html>
 
<style>
    .button {
    background-color: #008CBA;
    border: none;
    color: white;
    padding: 15px 32px;
    text-align: center;
    text-decoration: none;
    display: inline-block;
    font-size: 16px;
    margin: 4px 2px;
    cursor: pointer;
  }
</style>
 
<div id="chart_cpu_coral" style="margin: auto; width: 800px; height: 500px;"></div>


The Jetson Xavier power modes used were 2 and 6 (more information: [https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%2520Linux%2520Driver%2520Package%2520Development%2520Guide%2Fpower_management_jetson_xavier.html%23wwpID0E0OM0HA Supported Modes and Power Efficiency] )
<script>
      google.charts.load('current', {'packages':['corechart', 'bar']});
      google.charts.setOnLoadCallback(drawStuffCoralCpu);
     
      function drawStuffCoralCpu() {


== Benchmark results ==
        var chartDiv_Cpu_Coral = document.getElementById('chart_cpu_coral');


{| class="wikitable" style="display: inline-table;"
        var table_models_cpu_coral = google.visualization.arrayToDataTable([
! style="font-weight:bold; background-color:#efefef; color:#000000;" | Desktop PC
          ['Model',                      //Column 0
! colspan="2" style="text-align: center; font-weight:bold; background-color:#efefef; color:#000000;" | CPU Library
          'TensorFlow Lite \n Coral',
|-
          'TensorFlow Lite EdgeTPU \n Coral'],        //Column 1
| style="background-color:#e98d44; color:#000000;" | Model
          ['InceptionV1', 73, 32], //row 1
| style="background-color:#e98d44; color:#000000;" | Framerate
          ['InceptionV2', 72, 37], //row 2
| style="background-color:#e98d44; color:#000000;" | CPU Usage
          ['InceptionV3', 74, 14], //row 3
|-
          ['InceptionV4', 0, 5], //row 4
| style="background-color:#e98d44; color:#000000;" | Inception V1
          ['MobileNetV2',  0, 34], //row 5
| style="background-color:#fee3cd; color:#000000;" | 11.89
          ['MobileNetV2 + SSD',  0, 45]  //row 6
| style="background-color:#fee3cd; color:#000000;" | 48
        ]);
|-
        var Coral_materialOptions_cpu = {
| style="background-color:#e98d44; color:#000000;" | Inception V2
          width: 900,
| style="background-color:#fee3cd; color:#000000;" | 10.33
          chart: {
| style="background-color:#fee3cd; color:#000000;" | 65
            title: 'Model Vs CPU Load per backend',
|-
          },
| style="background-color:#e98d44; color:#000000;" | Inception V3
          series: {
| style="background-color:#fee3cd; color:#000000;" | 5.41
          },
| style="background-color:#fee3cd; color:#000000;" | 90
          axes: {
|-
            y: {
| style="background-color:#e98d44; color:#000000;" | Inception V4
              distance: {side: 'left',label: 'CPU Load'}, // Left y-axis.
| style="background-color:#fee3cd; color:#000000;" | 3.81
            }
| style="background-color:#fee3cd; color:#000000;" | 94
          }
|}
        };


{| class="wikitable" style="display: inline-table;"
        var materialChart_coral_cpu = new google.charts.Bar(chartDiv_Cpu_Coral);
! style="font-weight:bold; background-color:#efefef; color:#000000;" | Jetson Xavier (15W)
        view_coral_cpu = new google.visualization.DataView(table_models_cpu_coral);
! colspan="2" style="text-align: center; font-weight:bold; background-color:#efefef; color:#000000;" | CPU Library
! colspan="2" style="text-align: center; font-weight:bold; background-color:#efefef; color:#000000;" | GPU Library
|-
| style="background-color:#2c79d3; color:#000000;" | Model
| style="background-color:#2c79d3; color:#000000;" | Framerate
| style="background-color:#2c79d3; color:#000000;" | CPU Usage
| style="background-color:#2c79d3; color:#000000;" | Framerate
| style="background-color:#2c79d3; color:#000000;" | CPU Usage
|-
| style="background-color:#2c79d3; color:#000000;" | Inception V1
| style="background-color:#c5daf6; color:#000000;" | 8.24
| style="background-color:#c5daf6; color:#000000;" | 86
| style="background-color:#c5daf6; color:#000000;" | 52.3
| style="background-color:#c5daf6; color:#000000;" | 43
|-
| style="background-color:#2c79d3; color:#000000;" | Inception V2
| style="background-color:#c5daf6; color:#000000;" | 6.58
| style="background-color:#c5daf6; color:#000000;" | 88
| style="background-color:#c5daf6; color:#000000;" | 39.6
| style="background-color:#c5daf6; color:#000000;" | 42
|-
| style="background-color:#2c79d3; color:#000000;" | Inception V3
| style="background-color:#c5daf6; color:#000000;" | 2.54
| style="background-color:#c5daf6; color:#000000;" | 92
| style="background-color:#c5daf6; color:#000000;" | 17.8
| style="background-color:#c5daf6; color:#000000;" | 25
|-
| style="background-color:#2c79d3; color:#000000;" | Inception V4
| style="background-color:#c5daf6; color:#000000;" | 1.22
| style="background-color:#c5daf6; color:#000000;" | 94
| style="background-color:#c5daf6; color:#000000;" | 9.4
| style="background-color:#c5daf6; color:#000000;" | 20
|}


{| class="wikitable" style="display: inline-table;"
        function drawMaterialChart() {
! style="font-weight:bold; background-color:#efefef; color:#000000;" | Jetson Xavier (30W)
          var materialChart_coral_cpu = new google.charts.Bar(chartDiv_Cpu_Coral);
! colspan="2" style="text-align: center; font-weight:bold; background-color:#efefef; color:#000000;" | CPU Library
          materialChart_coral_cpu.draw(table_models_cpu_coral, google.charts.Bar.convertOptions(Coral_materialOptions_cpu));
! colspan="2" style="text-align: center; font-weight:bold; background-color:#efefef; color:#000000;" | GPU Library
|-
| style="background-color:#6aa758; color:#000000;" | Model
| style="background-color:#6aa758; color:#000000;" | Framerate
| style="background-color:#6aa758; color:#000000;" | CPU Usage
| style="background-color:#6aa758; color:#000000;" | Framerate
| style="background-color:#6aa758; color:#000000;" | CPU Usage
|-
| style="background-color:#6aa758; color:#000000;" | Inception V1
| style="background-color:#d8e9d3; color:#000000;" | 6.41
| style="background-color:#d8e9d3; color:#000000;" | 93
| style="background-color:#d8e9d3; color:#000000;" | 66.27
| style="background-color:#d8e9d3; color:#000000;" | 72
|-
| style="background-color:#6aa758; color:#000000;" | Inception V2
| style="background-color:#d8e9d3; color:#000000;" | 5.11
| style="background-color:#d8e9d3; color:#000000;" | 95
| style="background-color:#d8e9d3; color:#000000;" | 50.59
| style="background-color:#d8e9d3; color:#000000;" | 62
|-
| style="background-color:#6aa758; color:#000000;" | Inception V3
| style="background-color:#d8e9d3; color:#000000;" | 1.96
| style="background-color:#d8e9d3; color:#000000;" | 98
| style="background-color:#d8e9d3; color:#000000;" | 22.95
| style="background-color:#d8e9d3; color:#000000;" | 44
|-
| style="background-color:#6aa758; color:#000000;" | Inception V4
| style="background-color:#d8e9d3; color:#000000;" | 0.98
| style="background-color:#d8e9d3; color:#000000;" | 99
| style="background-color:#d8e9d3; color:#000000;" | 12.14
| style="background-color:#d8e9d3; color:#000000;" | 32
|}


== Framerate ==
          init_charts();
        }
        function init_charts(){
          view_coral_cpu.setColumns([0,1, 2]);
          materialChart_coral_cpu.draw(view_coral_cpu, Coral_materialOptions_cpu);
        }
        drawMaterialChart();
        }


[[File:Framerate Benchmarks gst-inference.png|1024px|frameless|thumb|center]]
</script>


== CPU Usage ==
</html>


[[File:CPU Benchmarks gst-inference.png|1024px|frameless|thumb|center]]
<noinclude>
{{GstInference/Foot|Example Applications/DispTec|Model Zoo}}
</noinclude>

Latest revision as of 20:39, 4 September 2024




Previous: Example Applications/DispTec Index Next: Model Zoo


GstInference
GstInference
Introduction
Getting started
Supported architectures

InceptionV1

InceptionV3

YoloV2

AlexNet
Supported backends

Caffe

Metadata and Signals
Overlay Elements
Utils Elements
Legacy pipelines
Example pipelines
Example applications
Benchmarks
Model Zoo
Project Status
Contact Us

Coral from Google




GstInference Benchmarks - Introduction

This wiki summarizes a series of benchmarks on different hardware platforms based on the run_benchmark.sh bash script that can be found in the official GstInference repository. The script is based on the following GStreamer pipeline: 

#Script to run each model
run_all_models(){

  model_array=(inceptionv1 inceptionv2 inceptionv3 inceptionv4 tinyyolov2 tinyyolov3)
  model_upper_array=(InceptionV1 InceptionV2 InceptionV3 InceptionV4 TinyYoloV2 TinyYoloV3)
  input_array=(input input input input input/Placeholder inputs )
  output_array=(InceptionV1/Logits/Predictions/Reshape_1 Softmax InceptionV3/Predictions/Reshape_1 
  InceptionV4/Logits/Predictions add_8 output_boxes )

  mkdir -p logs/
  rm -f logs/*

  for ((i=0;i<${#model_array[@]};++i)); do
    echo Perf ${model_array[i]}
    gst-launch-1.0 \
    filesrc location=$VIDEO_PATH num-buffers=600 ! decodebin ! videoconvert ! \
    perf print-arm-load=true name=inputperf ! tee name=t t. ! videoscale ! queue ! net.sink_model t. ! queue ! net.sink_bypass \
    ${model_array[i]} backend=$BACKEND name=net backend::input-layer=${input_array[i]} backend::output-layer=${output_array[i]} \
    model-location="${MODELS_PATH}${model_upper_array[i]}_${INTERNAL_PATH}/graph_${model_array[i]}${EXTENSION}" \
    net.src_bypass ! perf print-arm-load=true name=outputperf ! videoconvert ! fakesink sync=false > logs/${model_array[i]}.log
  done
}

Test benchmark video

The following video was used to perform the benchmark tests.
To download the video press right-click on the video and select 'Save video as' and save this on your computer.

Video 1. Test benchmark video

x86

The Desktop PC had the following specifications:

  • Intel(R) Core(TM) Core i7-7700HQ CPU @ 2.80GHz
  • 12 GB RAM
  • Linux 4.15.0-106-generic x86_64 (Ubuntu 16.04)
  • GStreamer 1.8.3

FPS Measurements

CPU Load Measurements

Jetson AGX Xavier

The Jetson Xavier power modes used were 2 and 6 (more information: Supported Modes and Power Efficiency)

  • View current power mode:
$ sudo /usr/sbin/nvpmodel -q
  • Change current power mode:
sudo /usr/sbin/nvpmodel -m x

Where x is the power mode ID (e.g. 0, 1, 2, 3, 4, 5, 6).

FPS Measurements

CPU Load Measurements

Jetson TX2

FPS Measurements

CPU Load Measurements

Jetson Nano

FPS Measurements

CPU Load Measurements

Google Coral

The following benchmarks were performed on the Coral Dev Board.

FPS Measurements

CPU Load Measurements


Previous: Example Applications/DispTec Index Next: Model Zoo



Retrieved from "https://developer.ridgerun.com/wiki/index.php?title=GstInference/Benchmarks&oldid=58370"