Google AutoML Vision-如何调用多个预测

时间:2018-09-07 22:02:06

标签: firebase google-cloud-platform google-vision automl google-cloud-automl

我正在基于Google I / O '17演示文稿创建一个Firebase托管的Web应用程序,其中Google的AutoML Vision可以拍摄云的图片,并根据其机器学习训练告诉您它是哪种云类型。我正在使用的代码只允许一个调用,我认为这是由于以下代码段造成的:

     // Get only the first prediction response
        let data = response[0]['payload'];
        predictions[data[0].displayName] = data[0].classification.score;

从Google文档中可以看出,[0]对应于annotateImageID。代码传递0的问题是,Web应用程序无法对多个图像进行预测。

下面是代码的整个部分,要求将结果推送到webapp:

exports.callCustomModel = functions.storage.object().onFinalize(event => {
const file = gcsClient.bucket(event.bucket).file(event.name);
let destination = '/tmp/' + event.name.replace(/\s/g, '');
return file.download({destination: destination})
    .then(() => {
        if(sizeOf(destination).width > 600) {
            console.log('scaling image down...');
            return resizeImg(destination);
        } else {
            return destination;
        }     
    })
    .then(() => {
        let bitmap = fs.readFileSync(destination);
        let data = new Buffer(bitmap).toString('base64');
        return callAutoMLAPI(data);  
    })
    .then((response) => {
        let predictions = {};

        // Get only the first prediction response
        let data = response[0]['payload'];
        predictions[data[0].displayName] = data[0].classification.score;

        if (Object.keys(predictions).length === 0) {
            predictions = {"predictionErr": "No high confidence predictions found"};
        }
        return db.collection('images').doc(event.name).set(predictions);
    })
    .then(() => {
        // Delete tmp image file to clean up resources
        return new Promise((resolve, reject) => {
            fs.unlinkSync(destination, (err) => {
            if (err) {
                reject(err);
            } else {
                resolve();
            }
            });
        });
    })

1 个答案:

答案 0 :(得分:0)

response.payload的所有预测都超出阈值集。