B-roll computational photography

For now, just adding interviews in autoEdit. What if could add also b-rolls/cutaways or obs doc scenes. and it recognises, objects, people etc.. computational photography capabilities.

Could add new column to paper-edit for visuals to go along side audio (transcription).

However not hat export would need to be as XML as EDL only supports one track. or could do 1 EDL per track. Which means two files. One for interview track and the other for video track.

For example here are some notes from the Watson conf 2016, but same could be done with Open source tech.

Visual recognition in video -draft notes

API key in process env variable

Ffmpeg select='not (mod 30 Every Number of frames Collect files as glob (node globe library) Spawn sync process node Loop through images Load image and send to visual recognition classifier Node A sync, img sent all at same time Code available on blog bluechasm Outliers generally 1 result only. -vsync cutting video without transcoding it.

Ffmpeg can pull out time stamp robert@bluechasm.com or tweet @RobertFromBC

http://blog.bluechasm.com/post/152964631846/turning-visual-recognition-into-video-recognition code example from blog:

//Included libraries
var glob = require('glob');
var watson = require('watson-developer-cloud');
var fs = require('fs-extra');
var sync = require('child_process').spawnSync;

//Visual Recognition API key setup
var visual_recognition = watson.visual_recognition({
  api_key: [YOUR API KEY HERE], // Demo Credentials.
  version: 'v3',
  version_date: '2016-05-20'

//Cut the video into frames: One every 30 frames
const procVid = sync('ffmpeg', ["-i" ,"video2.mp4","-vsync" ,"0", "-vf","select='not(mod(n,30))'","image_%d.jpg"]);

//Initialize the arrays and lists
var files= glob.sync('*.jpg' );
var x = files.length;
var j =0;
tags = []

//Tally up the results
function count(tags){
    var counts = {};
    for(var k = 0;k<tags.length;k++){
        counts[tags[k]] = (counts[tags[k]] || 0)+1;

//Iterate over the list of images
for(var i =0;i<x;i++){
    //Build request with the image you want to tag
    var params = {
        images_file: fs.createReadStream(files[i])

    //Send image to tag to the Classify endpoint
    visual_recognition.classify(params, function(err, res) {
        if (err){

        //Error Checking
        if(res.images[0].hasOwnProperty('error')||res.images[0]['classifiers'][0]['classes'].length ==0){
        //No errors: continue with storing tags
        else {
            console.log("Done Getting tags for: "+res.images[0].image);
        //Once you're done with the tags, count them up

Last updated