image-dataset/dist/client.js

Tool to build image dataset: collect, classify, review

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
const browser_1 = require("tensorflow-helpers/browser");
async function loadModels(args) {
    let baseModel = await (0, browser_1.loadImageModel)({
        url: '/saved_models/base_model',
        cacheUrl: 'indexeddb://base_model',
    });
    browser_1.tf.browser.toPixels;
    let classifierModel = await (0, browser_1.loadImageClassifierModel)({
        baseModel,
        modelUrl: '/saved_models/classifier_model',
        cacheUrl: 'indexeddb://classifier_model',
        checkForUpdates: true,
        classNames: args.classNames,
    });
    return classifierModel;
}
async function benchmark() {
    console.log('[benchmark] start');
    console.log('[benchmark] loading labels...');
    let labels;
    {
        let res = await fetch('/benchmark/labels');
        let json = await res.json();
        labels = json.labels;
        console.log({ labels });
    }
    console.log('[benchmark] loading images...');
    let images = [];
    {
        let res = await fetch('/benchmark/images');
        let json = await res.json();
        for (let url of json.images) {
            let image = new Image();
            await new Promise((resolve, reject) => {
                image.src = url;
                image.onload = resolve;
                image.onerror = () => {
                    reject(`Failed to load image: ${url}`);
                };
            });
            images.push(image);
        }
        console.log({ images });
    }
    console.log('[benchmark] loading models...');
    let classifierModel = await loadModels({ classNames: labels });
    console.log({ classifierModel });
    console.log('[benchmark] warm up model...');
    let n = 5;
    let i = 0;
    for (let image of images) {
        let result = await classifierModel.classifyImage(image);
        i++;
        document.body.textContent = `warm up ${i}/${n} images`;
        if (i >= 5) {
            break;
        }
    }
    console.log('[benchmark] classifying images...');
    document.body.textContent = `classifying ${images.length} images...`;
    n = images.length;
    i = 0;
    let startTime = performance.now();
    for (let image of images) {
        let result = await classifierModel.classifyImage(image);
        i++;
    }
    let endTime = performance.now();
    let duration = endTime - startTime;
    let seconds = duration / 1000;
    document.body.innerHTML = /* html */ `
  <p>classified ${n} images</p>
  <p>total time used: ${seconds}s</p>
  <p>images per second: ${n / seconds}</p>
  <p>ms per image: ${duration / n}</p>
  `;
    return {
        images_count: n,
        total_seconds_used: seconds,
        images_per_second: n / seconds,
        ms_per_image: duration / n,
    };
}
Object.assign(window, {
    loadModels,
    tf: browser_1.tf,
    cropAndResizeImageTensor: browser_1.cropAndResizeImageTensor,
    toTensor3D: browser_1.toTensor3D,
    benchmark,
});