Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Posenet model inconsistent results: Node vs Browser

See original GitHub issue

To get help from the community, check out our Google group.

I ran the posenet model on node 8.11.3 and I am comparing results of the node implementation for a given image with the browser implementation for the same image.

The results are inconsistent. Am I missing something? Or Is this to be expected?

Here is my node implementation.

global.XMLHttpRequest = require("xhr2");

const tf = require('@tensorflow/tfjs');
require('@tensorflow/tfjs-node');

const fetch = require('node-fetch');
const {Image, createCanvas} = require('canvas');
const posenet = require('@tensorflow-models/posenet')

async function run() {
  let img_path = 'ANY_IMAGE_URL';
  let buffer = await fetch(img_path).then(res => res.buffer());
  let img = new Image();
  img.src = buffer;
  const canvas = createCanvas(img.width,img.height);
  canvas.getContext('2d').drawImage(img,0,0);

  const imageScaleFactor = 0.5;
  const flipHorizontal = false;
  const outputStride = 8;
  const multiplier = 0.5;

  const net  = await posenet.load(multiplier);
  const pose = await net.estimateSinglePose(canvas, imageScaleFactor, flipHorizontal, outputStride);
  console.log(pose);
  return pose;
}

run();

NODE_RESULT:
{  
  "score":0.49880917983896594,
  "keypoints":[  
     {  
        "score":0.964009702205658,
        "part":"nose",
        "position":{  
           "x":584.1284123357551,
           "y":540.4772608240223
        }
     },
     {  
        "score":0.9792177677154541,
        "part":"leftEye",
        "position":{  
           "x":605.033953373647,
           "y":492.35327694003144
        }
     },
     {  
        "score":0.9352774024009705,
        "part":"rightEye",
        "position":{  
           "x":554.98182878691,
           "y":502.4654247774092
        }
     },
     {  
        "score":0.8619629740715027,
        "part":"leftEar",
        "position":{  
           "x":654.1466329203885,
           "y":516.7189273088338
        }
     },
     {  
        "score":0.12242487072944641,
        "part":"rightEar",
        "position":{  
           "x":518.3222664865168,
           "y":531.4258767239875
        }
     },
     {  
        "score":0.39839959144592285,
        "part":"leftShoulder",
        "position":{  
           "x":753.1191716727424,
           "y":696.705976944396
        }
     },
     {  
        "score":0.6442915797233582,
        "part":"rightShoulder",
        "position":{  
           "x":471.280373605403,
           "y":689.8485441580831
        }
     },
     {  
        "score":0.9180301427841187,
        "part":"leftElbow",
        "position":{  
           "x":895.1230138391418,
           "y":845.3166258401711
        }
     },
     {  
        "score":0.929760217666626,
        "part":"rightElbow",
        "position":{  
           "x":297.96139968722963,
           "y":814.5879274506809
        }
     },
     {  
        "score":0.20664989948272705,
        "part":"leftWrist",
        "position":{  
           "x":390.43972649092785,
           "y":970.6867716698673
        }
     },
     {  
        "score":0.3140019476413727,
        "part":"rightWrist",
        "position":{  
           "x":386.6281021896843,
           "y":972.8578296853178
        }
     },
     {  
        "score":0.19098210334777832,
        "part":"leftHip",
        "position":{  
           "x":472.2402253290729,
           "y":1026.7829016999826
        }
     },
     {  
        "score":0.19772598147392273,
        "part":"rightHip",
        "position":{  
           "x":518.5166831746552,
           "y":1029.4390766847068
        }
     },
     {  
        "score":0.19201675057411194,
        "part":"leftKnee",
        "position":{  
           "x":664.483786053818,
           "y":1456.2379157210196
        }
     },
     {  
        "score":0.21349774301052094,
        "part":"rightKnee",
        "position":{  
           "x":495.78668012422276,
           "y":1608.1637351606146
        }
     },
     {  
        "score":0.21133828163146973,
        "part":"leftAnkle",
        "position":{  
           "x":685.1607353550086,
           "y":1718.6341491358241
        }
     },
     {  
        "score":0.20016910135746002,
        "part":"rightAnkle",
        "position":{  
           "x":683.8384805983751,
           "y":1718.6952797660615
        }
     }
  ]
}

<!DOCTYPE html>
<html lang="en">
<head>
  <meta charset="UTF-8">
  <meta name="viewport"
        content="width=device-width, initial-scale=1.0">
  <meta http-equiv="X-UA-Compatible"
        content="ie=edge">
  <title>POSENET</title>
  <script src="https://unpkg.com/@tensorflow/tfjs"></script>
  <script src="https://unpkg.com/@tensorflow-models/posenet"></script>
</head>

<body>
<script>
   const multiplier = 0.5;
    const imageScaleFactor = 0.5;
    const flipHorizontal = false;
    const outputStride = 8;
    let estimator = null;
    
    function compute(src) {
      const imageElement = new Image();
      imageElement.src = src;
      imageElement.crossOrigin = "Anonymous"
      imageElement.onload = () => {
        const canvas = document.createElement('canvas');
        const ctx = canvas.getContext('2d');
        canvas.width = imageElement.width;
        canvas.height = imageElement.height;
        ctx.drawImage(imageElement, 0, 0, canvas.width, canvas.height)
        let imgdata = ctx.getImageData(0, 0, canvas.width, canvas.height);
        posenet
          .load(multiplier)
          .then(function (net) {
            if (!estimator) estimator = createEstimator(net);
            return estimator(imgdata, imageScaleFactor, flipHorizontal, outputStride)
          })
          .then((resp) => {
            console.log(JSON.stringify(resp));
          })
      }
    }

    function createEstimator(net) {
      return function (imageElement, scaleFactor, flipHorizontal, outputStride) {
        return net.estimateSinglePose(imageElement, scaleFactor, flipHorizontal, outputStride);
      }
    }  

    compute('SAME_IMAGE_URL')
</script>
</body>

</html>

BROWSER_RESULT:
{  
   "score":0.8681028520359713,
   "keypoints":[  
      {  
         "score":0.9993013143539429,
         "part":"nose",
         "position":{  
            "x":578.9772613558526,
            "y":531.6041130193785
         }
      },
      {  
         "score":0.9997183680534363,
         "part":"leftEye",
         "position":{  
            "x":611.7083526511923,
            "y":496.51715667554123
         }
      },
      {  
         "score":0.9996011853218079,
         "part":"rightEye",
         "position":{  
            "x":551.8440553083223,
            "y":502.5786637460719
         }
      },
      {  
         "score":0.9673418998718262,
         "part":"leftEar",
         "position":{  
            "x":658.4875402129563,
            "y":511.97225048555345
         }
      },
      {  
         "score":0.6721202731132507,
         "part":"rightEar",
         "position":{  
            "x":519.1532856530139,
            "y":512.3324056444221
         }
      },
      {  
         "score":0.8626165390014648,
         "part":"leftShoulder",
         "position":{  
            "x":732.3887470257787,
            "y":699.752749650838
         }
      },
      {  
         "score":0.9294714331626892,
         "part":"rightShoulder",
         "position":{  
            "x":476.7100266447285,
            "y":686.2188209235336
         }
      },
      {  
         "score":0.9612163305282593,
         "part":"leftElbow",
         "position":{  
            "x":899.471788768789,
            "y":843.1275232410964
         }
      },
      {  
         "score":0.9578098654747009,
         "part":"rightElbow",
         "position":{  
            "x":298.8155607706043,
            "y":823.8216600471369
         }
      },
      {  
         "score":0.8985893726348877,
         "part":"leftWrist",
         "position":{  
            "x":804.0089537447619,
            "y":991.5793307000699
         }
      },
      {  
         "score":0.9297114610671997,
         "part":"rightWrist",
         "position":{  
            "x":388.29281428996694,
            "y":990.1576139141062
         }
      },
      {  
         "score":0.7482747435569763,
         "part":"leftHip",
         "position":{  
            "x":673.7793992629657,
            "y":1070.5532308397346
         }
      },
      {  
         "score":0.7607749104499817,
         "part":"rightHip",
         "position":{  
            "x":521.7338706150117,
            "y":1047.567500109113
         }
      },
      {  
         "score":0.9209008812904358,
         "part":"leftKnee",
         "position":{  
            "x":665.2541383625241,
            "y":1438.8386407777584
         }
      },
      {  
         "score":0.754119873046875,
         "part":"rightKnee",
         "position":{  
            "x":511.0992839204372,
            "y":1440.0594120984638
         }
      },
      {  
         "score":0.7490795850753784,
         "part":"leftAnkle",
         "position":{  
            "x":674.7143572911896,
            "y":1719.4835948411314
         }
      },
      {  
         "score":0.6471004486083984,
         "part":"rightAnkle",
         "position":{  
            "x":508.224646022602,
            "y":1719.0399681389665
         }
      }
   ]
}

Issue Analytics

State:
Created 5 years ago
Comments:7 (3 by maintainers)

Top GitHub Comments

2reactions

jsumeetcommented, Aug 10, 2018

Any progress or updates on this issue? Were you able to reproduce this issue?

0reactions

rthadurcommented, Jun 11, 2022

Closing this issue as the tfjs-models/posenet has been deprecated in favor of tfjs-models/posenet-detection , please check the bennchmarktool to check the various backends here

Top Results From Across the Web

Comparing MoveNet to PoseNet for Person Fall Detection

The model returns the coordinates of the 17 key points along with a confidence score. MoveNet: MoveNet is the latest generation pose estimation ......

Human Pose Estimation Using TensorFlow's PoseNet Model

Is it possible to identify the movement/pose of a human body by doing an analysis of images and/or video? The answer to this...

There is no position detail ("x","y") in posenet TensorFlow ...

I ran the posenet model on node v8.11.0. Here is the code that I have run. the output results do not show the...

[Updated] BodyPix: Real-time Person Segmentation in the ...

When we released PoseNet last year (the first model ever that allowed for body estimation (what a Kinect does) in the browser using...

Nonintrusive Fine-Grained Home Care Monitoring - MDPI

The pose maps either past or present can be viewed from the browser of ... We used a PoseNet model for single-person pose...