Load and Pre-process Data

This topic describes how to load data into PerceptiLabs, pre-process it, and get to a working model:

• Select a Model Type

• Import a Dataset

• Define Your Data Settings

Select a Model Type

The first step towards creating a model in PerceptiLabs is to specify the model type:

Follow the steps below to specify a model type:

1. Click Create Project on the Overview screen to display the Load dataset popup:

2. Choose one of the follow model types:

Depending on your selection, PerceptiLabs will automatically customize the data settings if you import a public dataset. If you import a local dataset, you'll be able to perform the following:

a. Image Classification: You get to choose a top-level directory, where each sub-directory represents a class of images.

b. Segment Images: You get to choose two directories, one containing images and one containing masks.

c. Multi-Modal: For non-image classification/segmentation projects. Here, it's assumed that the data files live in the relative path(s) specified in the CSV file.

Note: Selecting Image Classification or Segment Images will also filter the list of public datasets that you can select from, to include only those suitable for the respective model type.

Import a Dataset

You can choose to:

• Import a Public Dataset
• Import a Local Dataset

Import a Public Dataset

This is a good choice for getting a model up and running quickly using existing, publicly-available data. Scroll to a public dataset, hover your mouse over it, and click Load:

PerceptiLabs will load your public dataset, allowing you to create models from it. Wait for the dataset to complete loading, and then click Create to create your first model:

Import a Local Dataset

Ensure you have a CSV file and the data it references, all on your local machine. Click Local at the top of the Data Wizard and click Upload .CSV :

Tip: Our Dataset Garden also has a number of curated datasets and CSV files that you can download and import to get you to a working model faster.

Depending on which model type you selected in above, you will be provided with different options:

Image Classification:

Click Browse, select the top-level directory containing your sub directories of images, and click Next:

Image Segmentation:

Click Browse for both of the folder selections to specify the directories for the images and masks respectively, and then click Next to continue:

Multi-Modal:

Click Browse, select the CSV to upload, and click Next:

Define Your Data Settings

After you have imported a dataset, you must the define your data settings:

1. Define your Data Settings so that the Input and Target are on the correct columns:

Note: The name you specify for the model will become the name of the directory where the model.json file is stored. The base path defaults to ~/Users/<username>/Documents/PerceptiLabs/Default.

2. (Optional) Click on the pre-processing button in the column heading to configure data pre-processing settings:

Note: This button is only available for certain datatypes (e.g., images, numerical data, etc.)

This allows you to define any pre-processing steps that PerceptiLabs should perform on your data (e.g., resizing images):

For a list and description of each pre-processing setting, see the Pre-processing Settings reference guide.

After configuring your pre-processing settings, click Save to return to the Data Wizard.

3. Click Create. PerceptiLabs will generate a model and navigate you to the Modeling Tool.

Note: Upon completing the Data Wizard, PerceptiLabs will create a directory with the model.json file on disk, and the model will be listed on the Overview screen. Any time you save the model, PerceptiLabs will update the model's model.json file.

Congratulations on having created your first project and model in PerceptiLabs! Make sure you read Build Models to learn how to work with the model.