​

🚀 Quickstart Guide

Welcome to Ocular AI! This guide will help you get started with our end-to-end platform for creating specialized computer vision models. Follow these steps to begin transforming your unstructured image and video data into powerful AI models.

Before diving in, make sure you’ve completed the Setup Guide to create an organization and workspace. Once you’re set up, here’s how to get started:

​

1. Data Ingestion and Curation

First, let’s get your data into the platform:

1. Navigate to the Ingestion tab on the left sidebar
2. Choose from our available storage integrations:
   • AWS S3 Buckets
   • Azure Storage
   • Google Cloud Platform Storage
   • Local file upload

Use our File Explorer to easily browse and select data from your connected storage sources. Follow the prompts to ingest data into the Catalog.

Once your data is in the Catalog, you can use the Search tab to explore and filter your data and create Indexes to group related data together.

Indexes are powerful, reusable collections that help you organize files based on common characteristics or use cases. You can leverage these Indexes across multiple projects to maintain consistency in your workflows.

​

2. Create a Project

Before starting annotations, create a project to organize your work:

1. Click on the Projects tab on the left sidebar, then the New Project button
2. Choose a descriptive name and optional description and create the project.
3. Once created, configure the project:
   • Create labels
   • Invite teammates and annotators

Your project serves as a central workspace where you’ll manage annotations, model training, and evaluation workflows.

​

3. Data Annotation

Once you have a project and an index, you can start annotating:

​

Using the Kanban Board

The Kanban board organizes your annotation workflow into columns:

• Batches: Available image groups ready for annotation. To create a batch, follow the prompts to ingest data from an Index.
• Annotating: Active annotation jobs, created from batches.
• Review: Completed jobs awaiting review
• Dataset: Approved frames organized into train/valid/test sets
• Rejected: Annotation jobs that didn’t meet quality standards

​

Annotation Methods

Choose from multiple annotation approaches:

1. Manual Annotation

   • Use our Infinite Canvas with bounding box or polygon tools, including the Smart Polygon tool with SAM2-powered auto-segmentation
   • The Canvas Agent (AI-assisted annotation) can help to speed up this process

2. Auto Annotation

   • Create fully auto-labeled jobs
   • Review and adjust auto-generated annotations

​

4. Dataset Management

Organize your annotated data effectively. Check out the Dataset and Versions tabs, where you can:

1. Create dataset versions to track changes
2. Split and adjust datasets into training, validation, and test sets
3. Export datasets in various formats for model training

​

5. Model Training

Train custom models using our integrated AI capabilities:

1. Navigate to a specific version and click the Train button to train a model from that bersion
2. Configure your training job:
   • Select model variant (n/s/m/l/x)
   • Set epochs and image size
3. Monitor training progress and metrics

​

6. Model Evaluation

After a training run has completed, navigate to the Models tab to see details about each model training run:

1. Review training metrics and graphs
2. Analyze confusion matrices
3. Observe tests on validation data

​

7. Integration Options

Connect Ocular with your workflow:

1. API Access

   • Use our REST API for direct integration
   • Secure authentication with API keys

2. SDK Integration

   • Install our Python SDK: pip install ocular-ai
   • Programmatically access datasets and models

3. Jupyter Notebooks

   • Use our pre-built notebooks for:
     • Dataset downloading
     • Training job management
     • Custom evaluation plots

​

Next Steps

• Explore our detailed documentation for in-depth guides
• Join our community for support and best practices
• Check out example projects and use cases