How To Easily Build Your First AI Computer Vision Project On Tuba.AI

The world of AI computer vision can seem daunting, especially for those without extensive coding experience.  However, the desire to digitise your workflow or business should not be hindered by technical barriers.  Well, what if there was a single tool that could simplify the process?

Enter Tuba.AI.

This one-stop platform empowers anyone to build AI computer vision applications, eliminating the need to write a single line of code thanks to its intuitive No-Code interface. Offered as a Software-as-as-Service (SaaS), Tuba.AI is the perfect user-friendly platform that streamlines the entire AI computer vision process, guiding users through dataset labelling, model training and deployment. 

But what truly sets Tuba.AI apart is its modular approach. This means you can tackle each step independently, offering maximum flexibility for your specific needs.

Therefore, in this article, we will take you on a step-by-step journey through building an AI computer vision model on Tuba.AI. By exploring each task individually, we will showcase the power and convenience of this platform’s modularity. But before we dive in, be sure to sign-up to Tuba.AI here.

Dataset Labelling:

If you are embarking on your AI computer vision journey from the ground up, then dataset labelling is the first crucial step. When creating a new project on Tuba.AI, you will need to specify the name of the project, as well as providing a concise description about it. Additionally, ensure you select “Labelling” as the project type before proceeding.

Creating a “Labelling” Project on Tuba.AI

Once you have defined your project, Tuba.AI will guide you to the data labelling method selection page. Here, you will choose the most suitable approach for your model from three options:

Classification: Ideal for categorising objects within your data.

Detection: Perfect for pinpointing and identifying the location of specific objects within your data.

Segmentation: Meant for outlining the precise boundaries of objects in your data.

Selecting a “Labelling Type” on Tuba.AI

After selecting your data labelling method, it’s time to define the labels you’ll use for your project. Simply enter each label name separated by a comma (e.g. apple, banana, orange).

Adding labels for your project on Tuba.AI

Now it is time to upload your dataset to Tuba.AI! For your convenience, Tuba.AI offers the option of uploading your dataset directly from your PC, or from your Google Drive account.

Uploading datasets on Tuba.AI

The image below demonstrates your project’s central hub. Here, you will find all your uploaded image datasets neatly organised for easy labelling. Tuba.AI provides real-time progress tracking, giving you a clear picture of your labelling completion rate. This empowers you to effectively manage your project, ensuring data quality and consistency. Once you have meticulously labelled your data, simply click the export button to effortlessly download it.

Tuba.AI’s central hub for labelling projects

Case: Classification Labelling

When it comes to labelling images through classification, you will essentially assign a single category label to the entire image. For example, the image below might be labelled as “crack.”

Labelling via “Classification” on Tuba.AI

Case: Detection or Segmentation Labelling

As for detection or segmentation, the labelling process is slightly different. In these cases, you can pinpoint and label multiple objects within an image or define the exact boundaries of objects with pixel-level detail. This expanded labelling power makes detection and segmentation ideal for tasks requiring a more nuanced understanding of your image data. For example, with segmentation, you could specifically identify the “Crack” parameters on the glass image below.

Labelling via “Segmentation” on Tuba.AI

For segmentation and detection tasks, keep in mind that larger datasets (ideally hundreds of images) tend to yield higher accuracy.

Model Training:

Train your AI model on Tuba.AI regardless of where your labelled dataset originated from. Whether you meticulously labelled it on another platform, or on Tuba.AI, you can seamlessly import it for model training. This is the power of Tuba.AI’s modular design – you can leverage your existing efforts!

Simply create a new project on Tuba.AI. Give it a clear name and description, and choose “Training” as your project type. And let us get started on building your powerful AI computer vision model!

Creating a “Training” project on Tuba.AI

Training on Tuba.AI is a breeze. Simply upload your labelled dataset either directly from your PC or your Google Drive account. Before we proceed, you must confirm the labelling method used for your model (classification, detection, or segmentation). This helps Tuba.AI optimise the training process for the best results.

Uploading a Dataset for Training on Tuba.AI

Onto the final stage, you will get to choose how you want to configure your model. You can either fine-tune it manually, or leverage Tuba.AI’s powerful AutoML engine for automatic configuration, ensuring the best possible model for your specific needs. Once you hit “Start Training,” sit back and relax! Tuba.AI will keep you informed via email about the training progress and successful completion of your AI computer vision model.

Selecting Training Configuration Type on Tuba.AI

Model Deployment:

Regardless of where you trained your AI computer vision model, Tuba.AI seamlessly handles deployment for real-world application. Simply create a new project, name it clearly, describe its purpose, and choose “Model Deployment Only” as the project type. Tuba.AI bridges the gap, allowing you to leverage your existing model for immediate impact.

Creating a “Model Deployment Only” project on Tuba.AI

On the next page (as shown below), you will help Tuba understand your model by specifying the labelling method used (classification, detection, or segmentation) and listing the different labels you assigned to your data (separated by commas). Finally, upload your trained model directly from your PC or your Google Drive.

Uploading a Model for Deployment on Tuba.AI

Once deployed, go to the “Home” page, then head to the Tuba.AI’s “Dashboard” and select your newly deployed model. With a simple click on “Predict,” you can test your AI computer vision model and assess its accuracy. Tuba.AI makes it easy to validate your hard work and ensure your model performs as expected!

Testing Models on Tuba.AI

Sign-up to Tuba.AI now to start your free trial!

If you are looking to personalise AI computer vision applications to your specific needs, explore Tuba.AI’s polished suite of Software Development Kits (SDKs). These powerful tools empower you to craft custom AI computer vision applications that seamlessly integrate into your existing workflows. Tuba.AI’s SDKs include: Automatic Image Labelling, Classification Model Training, Object Detection or Segmentation Training, Model Deployment, and Job Manager. With Tuba.AI’s SDKs, the possibilities for transforming your business with AI computer vision are endless.

You can request Tuba.AI’s SDKs here.

Important Announcements:

Don’t miss out on a unique opportunity to uncover the latest AI trend, large vision models (LVMs), by tuning in on our 3-part webinar series for free! 

The first webinar, ‘Large Vision Models: Theory and Applications’, goes live on Friday, May 10th, 2024 at 2:00 PM CET (3:00 PM GMT+3), and it will be hosted by our Co-founder and Tech Lead, Mohamed Rashad!

Register now!


Leave a Reply

Your email address will not be published. Required fields are marked *