This tool loads a Mobilenet image classification model into RedisAI, takes a bunch of images that you provide on the command line and classifies them with it.
-
First and foremost, you need to have Redis and RedisAI installed. You can build this all from source, but I find it a lot easier to just use Docker. In the root of this project, I have a
start-redis.shfile. It will download the needed image, make sure RedisAI is configured, and have you up and running in a jiffy (or is that giphy).$ ./start-redis.sh
-
Also, make sure you have Node installed. I like to use Node Version Manager but do it how you like.
-
Clone this repo.
$ git clone git@github.com:redislabs-training/redisai-node-example.git
-
Install all of the dependencies.
$ npm install
-
There is no step, five. You're all set!
This is a simple example that uses the CLI. Everythign you need should be included in the repo. The model is in the model folder. There are even some images to play with in the img folder.
To user the tool to classify an images, just run:
$ npm start img/panda.jpg
It'll return the top 5 most likely classes for the images, including their scores:
┌─────────┬──────────────────────────────────┬───────────────────────┐
│ (index) │ label │ score │
├─────────┼──────────────────────────────────┼───────────────────────┤
│ 0 │ 'giant panda' │ 0.9475338459014893 │
│ 1 │ 'Staffordshire bullterrier' │ 0.0028441373724490404 │
│ 2 │ 'American Staffordshire terrier' │ 0.0027995011769235134 │
│ 3 │ 'French bulldog' │ 0.002384455408900976 │
│ 4 │ 'badger' │ 0.0011667307699099183 │
└─────────┴──────────────────────────────────┴───────────────────────┘
img/panda.jpg
The CLI is variadic, so you can add multiple images as well:
$ npm start img/panda.jpg img/sample_dog.jpg img/sample_computer.jpg
┌─────────┬──────────────────────────────────┬───────────────────────┐
│ (index) │ label │ score │
├─────────┼──────────────────────────────────┼───────────────────────┤
│ 0 │ 'giant panda' │ 0.9475339651107788 │
│ 1 │ 'Staffordshire bullterrier' │ 0.002844122936949134 │
│ 2 │ 'American Staffordshire terrier' │ 0.002799485344439745 │
│ 3 │ 'French bulldog' │ 0.0023844465613365173 │
│ 4 │ 'badger' │ 0.0011667292565107346 │
└─────────┴──────────────────────────────────┴───────────────────────┘
img/panda.jpg
┌─────────┬───────────────────┬─────────────────────┐
│ (index) │ label │ score │
├─────────┼───────────────────┼─────────────────────┤
│ 0 │ 'malamute' │ 0.3466069996356964 │
│ 1 │ 'Eskimo dog' │ 0.24618299305438995 │
│ 2 │ 'Siberian husky' │ 0.15509867668151855 │
│ 3 │ 'dogsled' │ 0.03109908662736416 │
│ 4 │ 'German shepherd' │ 0.00960743147879839 │
└─────────┴───────────────────┴─────────────────────┘
img/sample_dog.jpg
┌─────────┬─────────────────────┬─────────────────────┐
│ (index) │ label │ score │
├─────────┼─────────────────────┼─────────────────────┤
│ 0 │ 'screen' │ 0.29382583498954773 │
│ 1 │ 'desktop computer' │ 0.164292573928833 │
│ 2 │ 'computer keyboard' │ 0.08968286216259003 │
│ 3 │ 'mouse' │ 0.06009386107325554 │
│ 4 │ 'monitor' │ 0.05255772918462753 │
└─────────┴─────────────────────┴─────────────────────┘
img/sample_computer.jpg
And, it will even take a URL to an image, like this rando I grabbed off the Internet:
$ npm start http://guyroyse.com/guy-royse-papis-global-2018.jpg
Apparently, I'm a fur coat.
┌─────────┬──────────────────────┬──────────────────────┐
│ (index) │ label │ score │
├─────────┼──────────────────────┼──────────────────────┤
│ 0 │ 'fur coat' │ 0.3993118405342102 │
│ 1 │ 'cellular telephone' │ 0.049747321754693985 │
│ 2 │ 'stole' │ 0.03721235692501068 │
│ 3 │ 'wool' │ 0.03711461275815964 │
│ 4 │ 'wig' │ 0.02954762428998947 │
└─────────┴──────────────────────┴──────────────────────┘
http://guyroyse.com/guy-royse-papis-global-2018.jpg