Miss any of our Open RFC calls?Watch the recordings here! »

mean-average-precision

1.7.0 • Public • Published

Build Status

mean-average-precision

For explanation about mAP, please read https://github.com/Cartucho/mAP

install

npm install mean-average-precision

prediction format

For bounding box mAP, every prediction/ground truth objects should look like

{
    label: "car",
    filename: "image1.jpg",
    left: 22,
    top: 34,
    confidence: 0.9, // only for predictions
    right: 231,
    bottom: 78,
}

Basic Example

Simple average precision score

mAP.apScore( [{
        gt: 0, 
        confidence: 0.1
    },{
        gt: 0, 
        confidence: 0.4
    },{
        gt: 1, 
        confidence: 0.35
    },{
        gt: 1,
        confidence: 0.8
}]);
// => 0.833333

Mean average precision on bounding box predictions vs groundtruth

By default, it will compute the AP[iou=0.5]

const mAP = require('mean-average-precision')
 
const groundTruths = [{
            filename: "image1.jpg",
            label: "car",
            left: 22,
            top: 34,
            right: 231,
            bottom: 78,
        },{
            filename: "image1.jpg",
            label: "pedestrian",
            left: 22,
            top: 34,
            right: 231,
            bottom: 78,
    }];
 
const predictions = [{
        filename: 'image1.jpg',
        confidence: 0.9,
        label: 'car',
        left: 25,
        top: 38,
        right: 201,
        bottom: 90
    }, {
        filename: 'image1.jpg',
        label: 'pedestrian',
        confidence: 0.7,
        left: 32,
        top: 39,
        right: 452,
        bottom: 92
    }, {
        filename: 'image1.jpg',
        confidence: 0.5,
        label: 'car',
        left: 541,
        top: 42,
        right: 621,
        bottom: 94
    }];
    
    
mAP({
    groundTruths,
    predictions
});

Set iouThreshold

// IoU Threshold default value is 0.5
// you can change it using iouThreshold
 
mAP({
    groundTruths,
    predictions,
    iouThreshold: 0.6
});

List misclassified predictions

// You can have more information 
// on which predictions are considered as
// misclassified by doing
 
mAP.listMisclassified({
    groundTruths,
    predictions,
    iouThreshold: 0.6
});

IoU

// You can use the IoU function directly using
 
mAP.iou({
    left: 22,
    top: 34,
    bottom: 38,
    right: 30
},{
    left: 21,
    top: 32,
    bottom: 40,
    right: 32
});

Other distance metrics

You can use mean-average-precision implementation with other distance metrics by using

// You can run it with for example an euclidian distance
 
mAP({
    groundTruths,
    predictions,
    dist: (pred, groundTruth) => Math.sqrt(((pred.x - groundTruth.x) * (pred.x - groundTruth.x)) + ((pred.y - groundTruth.y) * (pred.y - groundTruth.y))),
    threshold: 10
});

Install

npm i mean-average-precision

DownloadsWeekly Downloads

61

Version

1.7.0

License

MIT

Unpacked Size

28.6 kB

Total Files

17

Last publish

Collaborators

  • avatar