generics.js/src/Convolution.js

A minimal library for Deep learning for the web

var { Image } = require('image-js');

function make_mat(data){
    var res=[];
    var ctr=0;
    for(var i=0;i<Math.ceil(Math.sqrt(data.length))-1;i++){
        var row=[];
        for(var j=0;j<Math.ceil(Math.sqrt(data.length))-1;j++){
            row.push(data[ctr]);
            ctr++;
        }
        res.push(row);
    }
    return res;
}

function relu(arr){
    //max(0,num);
    for(var i=0;i<arr.length;i++){
        arr[i]=Math.max(0,arr[i]);
    }
    return arr;
}
function max_pooling(raw_arr,kernel_length){
    var arr=Array.from(raw_arr);
    Array.prototype.max = function() {
        return Math.max.apply(null, this);
    };
    var res=[];
    if(arr.length<kernel_length){
        throw new Error("kernel is bigger then the image ");
    }else{
        var i,j,temparray,chunk = kernel_length;
        for (i=0,j=arr.length; i<j; i+=chunk) {
            temparray = arr.slice(i,i+chunk);
            var pooled=Math.max.apply(null,temparray);
            res.push(pooled/255);
        }
        return res;
    }
}


async function image_flattenX(dir,img_length,img_height,kernel_size,conv_options,conv_kernel) {
    var flatten=[];
    var img = await Image.load(dir);
    var image=img.resize({ width:img_length,height:img_height});
    img=img.gray();
    if(conv_kernel!=null){
        image=image.convolution(conv_kernel,conv_options);
    }

    flatten.push(max_pooling(relu(image.data),kernel_size));

    try{
        process.stdout.cursorTo(0);
        process.stdout.write("Processed the image with dir: "+dir);
    }catch(err){
        console.info("Processed the image with dir: "+dir);
    }

    return flatten;


}

/**
* A class to process the image 
*/
class Convolution{

            /**
    *  To process the image - Resize,convolution and flatten a image. 
    * @param {string} directory  - The amount of neurons for each layer.
    * @param {number} img_length  - The activation function for each layer.
    * @param {number} img_height  - Accepts the JSON data type.
    * @param {number} kernel_size  - Accepts the JSON data type.
    * @param {number} conv_options  - Accepts the JSON data type.
    * @param {array} conv_kernel  - Accepts the JSON data type.
    *@returns {array} Image in array format.
    */
    async image_flatten (dir,img_length,img_height,kernel_size,conv_options,conv_kernel){
        return image_flattenX(dir,img_length,img_height,kernel_size,conv_options,conv_kernel);
    }

}

module.exports = Convolution;