Search Unstructured Video

Use Computer Vision and Deep Learning to search untagged video

Search our video database for an object

Examples: chair, couch, rabbit, elephant

How it works

Using computer vision to search unstructured video

Tools used

Source Data

This demo uses videos pulled directly from YouTube, but can easily be modified to handle any video source by either downloading them directly from a remote source, or by using the Data API to connect to Dropbox or Amazon S3.

Processing steps

  • 1. Download video

    The video file is downloaded from YouTube, Dropbox, S3, or another source, and its file format is examined before proceeding.

  • 2. Split in frames

    The Video Metadata Extraction algorithm splits the video into frames, then processes them in batches.

  • 3. Auto-generate frame tags

    Tags and annotations describing each frame are generated using InceptionNet and other image classification/extraction algorithms.

  • 4. Search tags

    After the video is processed, timestamps are assigned to each extracted tag, allowing users to search across one or many videos for specific tags.


With just a few dozen lines of code and a couple spare hours, it's possible to build an entire video processing and tagging pipeline on top of Algorithmia's tools. Add in a few other relevant algorithms such as Approximate Nearest Neighbors Clustering, Image Similarity, or Deep Filter, and you can quickly build a video recommender engine, copy detector, or filter.

Built For Developers

A simple, scalable API for machine intelligence


import Algorithmia
input = {
  "input_file": "data://path/to/file.mp4",
  "output_file": "data://save/data.json",
  "algorithm": "algo://deeplearning/IllustrationTagger/0.2.5"
client = Algorithmia.client('API KEY')
algo = client.algo('media/VideoMetadataExtraction/0.1.5')
print algo.pipe(input)


  "output_file": "data://save/data.json"

Join the thousands of developers already building intelligent apps

Get 10,000 additional credits when you sign up using the code "demos"


Algorithms as a Microservice

Leverage an ever-growing library of more than 2,800 algorithmic microservices via an intuitive API. We provide the tools and manage the cloud infrastructure needed to run it at scale.

Learn more

Web Services for Business Logic

Instantly deploy your backend code as an API for public or private consumption. Every algorithm runs as it's own microservice, making each composable, interoperable, and secure.

Learn more

Hosted Trained Models

Have a trained machine learning or deep learning model? Turn it into a serverless microservices in minutes. We'll show you how to get started for free, and scale with ease.

Learn more