Welcome to the clevrML Demo. In this demo, you will experience what it is like to make an AI model on the clevrML platform. 

The way AI learns to do a task is by showing it examples of whatever task you want to solve with answers next to the examples. The examples are called our dataset and answers are called LabelsTypically, you will call a group of data with the same label as a class.

 

Let's take a look at our dataset with the corresponding labels:

Dataset

Image by John Matychuk

Stop Sign

traffic-light3.jpg

Traffic Light

bicycle3.jpg

Bicycle

traffic_light3.jpg

Traffic Light

bicycle2.jpg

Bicycle

stop_test2.jpg

Stop Sign

traffic-light2.jpg

Traffic Light

stop_test6.jpg

Stop Sign

bicycle1.jpg

Bicycle

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

from clevrml import

Image_Model

import os

key = os.environ[ "API-KEY" ]

model = Image_Model( )

model.build_model(

   api_key=key,

   class_names=

   example_folders=

   model_name=

)

<Pending Input>,

<Pending Input>,

<Pending Input>,

Output

Your model has successfully been built!

Details:

-------------------------------------

Model Name: 

Deployed: True

Build: Complete

Cost: $5.45

------------------------------------

JSON: 

{

   Model_Name:

   Deployed: True,

   Build: "Complete",

   Cost: 5.45

}

mark

mark

This is our dataset that will help our AI model learn how to recognize various traffic objects. The images are the actual data and the name below is the labels for the image. 

This may seem very obvious for humans, but for computers, this isn't as easy. We need to help our model by providing the labels with the corresponding image so that it can generalize for later. 

One advantage of using clevrML is the amount of data needed to build AI models. clevrML has built a world-class technology called "Active Memory Learning" which allows you to supply as little as three examples per "class" (a group of labels) to generalize well in a task. Contrast this with Neural Networks (a popular AI method) would require ~10,000 images of Stop Signs, traffic lights and Bicycles (a total of 30,000 images in a dataset).

Let's start building our model. We need to first make classes (groups) for Traffic Lights, Stop Signs and Bicycles. When using the clevrML Image Model API, classes need to be in a folder with the corresponding data. For this demo however, this is all been done for you (ie: The folder on the right side of the screen "/Home/Traffic-Light-Data/"

Let's start with Traffic Lights. Click on all the Traffic Light images to make our "Traffic Light" class.

Hint: Click on the images with a yellow border.

folder_edited.png

/Home/Traffic-Light/

Now we need to build the Stop Sign class. Like the previous step, click on all the images of Stop Signs to add the images to our Stop Sign folder.

Hint: Click on all the images with a yellow border.

Finally, let's add the remaining data to our Bicycle class. Click all the images of Bicycles to add the images to the folder.

Hint: Click all of the images with a yellow border.

Typically when using the clevrML SDK, you need to provide the model with the names of the classes so the model knows what to tell you is being predicted. For this demo, this step has been done for you automatically. 

We need to give our model a name. For this step, pick any name you would like.

Our code is built and ready to use. Click the "Run Code" button below to send the request to the Image Model API

loading.gif

Our model has been built using the Image Model API. What you see is the console output from clevrML's servers telling us all the information we need.

 

Typically, the building process can take hours, days or even weeks in runtime with Neural Networks. On clevrML however, building a model takes seconds thanks to Active Memory Learning.

Before starting, what is this code? This is a snippet of the official Python Software Development Kit (SDK) for clevrML. The purpose of an SDK is to make APIs easily accessible for the developer by providing ready-made code that can easily communicate with our servers in the cloud (ie: where all AI models on clevrML are built and deployed)

For the purpose of this demo however, you do not need to know any code. In this demo, you will use an easy UI to build this code snippet to use the clevrML API.