zqingr / tfjs-yolov3 Goto Github PK

View Code? Open in Web Editor NEW

83.0 3.0 16.0 12.04 MB

A Tensorflow js implementation of YOLOv3 and YOLOv3-tiny

Home Page: https://zqingr.github.io/tfjs-yolov3-demo/

HTML 5.02% TypeScript 66.02% CSS 1.78% JavaScript 27.19%

tfjs-yolov3's Introduction

中文 | English

tfjs-yolov3

介绍

完全用js来实现图片中的目标检测基于yolov3算法和Tensorflow.js库用tensorflow.js实现yolov3和yolov3-tiny

需要注意的是: 必须是[email protected]版本以上

特点

可以识别任意尺寸的图片
同时支持yolov3和yolov3-tiny

快速开始

安装

npm install tfjs-yolov3

用法示例

import { yolov3, yolov3Tiny } from 'tfjs-yolov3'

async function start () {
  const yolo = await yolov3Tiny() // pre-load model (35M)
  // or
  // const yolo = await yolov3() // pre-load model (245M)

  const $img = document.getElementById('img')
  const boxes = await yolo($img) 
  draw(boxes) // Some draw function
}
start()

API 文档

yolov3和yolov3Tiny函数接受一个options对象，并返回一个函数

export declare function yolov3 (
  { modelUrl, anchors }? :
  { modelUrl?: string, anchors?: number[] }
): Promise<yolo>

export declare function yolov3Tiny (
  { modelUrl, anchors }? :
  { modelUrl?: string, anchors?: number[] }
): Promise<yolo>

参数	说明
modelUrl	可选，预训练model的url，可把model下载到本地，加快预训练model的加载速度，点我下载
anchors	可选，可自定义anchores，格式参考config

这两个函数调用后会加载预训练model，并返回一个函数，可用这个函数去识别图片，并返回识别后的box列表，参数如下：

type yolo = ($img: HTMLImageElement) => Promise<Box[]> 

interface Box {
  top: number
  left: number
  bottom: number
  right: number
  width: number
  height: number
  scores: number
  classes: string
}

DEMO

点击查看在线DEMO

tfjs-yolov3's People

Contributors

Stargazers

Watchers

Forkers

guyperl hpssjellis gbrian dimitrius-ion djabdou50 leoxing1996 fightyang dorucioclea wjlee-barco dkrivec zhaoyk1986 thesamethe gary109 huhaorui liyunfei0411 jef-way

tfjs-yolov3's Issues

npm install 之后怎么启动呢

node_module issue

After installing using npm i tfjs-yolov3

I get this error from importing it into my project. util.js does exist.

node_modules\@tensorflow\tfjs-data\dist\util\deep_map.js: util could not be found within the project or in these directories:
  node_modules\tfjs-yolov3\node_modules
  node_modules

If you are sure the module exists, try these steps:
 1. Clear watchman watches: watchman watch-del-all
 2. Delete node_modules and run yarn install
 3. Reset Metro's cache: yarn start --reset-cache
 4. Remove the cache: rm -rf /tmp/metro-*
  54 | Object.defineProperty(exports, "__esModule", { value: true });
  55 | var tf = require("@tensorflow/tfjs-core");
> 56 | var util_1 = require("@tensorflow/tfjs-core/dist/util");
     |                                                  ^
  57 | var util_2 = require("util");

请问我的anchor参数，和labels是如下的，其中anchor是二维数组的这种，不是一维的，要怎么用您的这个项目

"labels" : [
    "hololens"
],
"anchors" : [
    [160, 171, 261, 167, 301, 285],
        [78, 86, 104, 69, 138, 105],
        [37, 32, 58, 61, 62, 44]
]

请问，怎么启动，给给教程可以吗？

yarn start 启动之后，报错

还有，能再给个打包的命令吗？npm run build，yarn build 都打不上包

能写个react.js版本吗？实在搞不定webpack和这些依赖，老是报错

About the website problem

Yolov3 can't work properly. What happened to it?(Yolov3-tiny seems good though)

load yolo custom model

Hello
I created a model.weights from yolov3
I wonder if it is possible to load it to work ?

thank you

404 on https://cdn.jsdelivr.net/npm/@tensorflow/tf.min.js.map

Hi @zqingr

Thank you for the project, very cool stuff!!!

I get 404 on https://cdn.jsdelivr.net/npm/@tensorflow/tf.min.js for some reason.
should it be https://cdn.jsdelivr.net/npm/@tensorflow/tfjs/dist/tf.min.js?

Would like to contribute or collaborate if you are looking for help.

Dimitrius

how to use

Document.html:341 Uncaught (in promise) TypeError: yolov3Tiny is not a function
at start (Document.html:341)
at Document.html:350

i had this question,i use it in electron
i only use your index.bundle.js to run ok
我无法使用如下的方式去运行：
用法示例
import { yolov3, yolov3Tiny } from 'tfjs-yolov3'

async function start () {
const yolo = await yolov3Tiny() // pre-load model (35M)
// or
// const yolo = await yolov3() // pre-load model (245M)

const $img = document.getElementById('img')
const boxes = await yolo($img)
draw(boxes) // Some draw function
}
start()

似乎存在内存泄露的情况

在最新版本Edge浏览器(100.0.1185.44)，Windows 11 22589.100 系统下，连续识别多张图片，会导致GPU进程占用大量内存，且随着图片数量的增加而增加。怀疑存在内存泄漏的情况。