Preparing save…

EXPERIMENT 02 · AI VISION

Teach a computer to recognise what it sees.

Give the machine examples, train it, then test whether it can recognise a new image.

01

FIRST VISION MISSION

Teach the computer to recognise fruit

Create classes such as Mango, Orange and Banana, add different examples, train the model and test a new image.

CREATE→ADD EXAMPLES→TRAIN→TEST→UNDERSTAND
1

CREATE + ADD EXAMPLES

Build your training set

Each class is a labelled category you want the computer vision model to recognise. Start with sample images, upload your own images or use the camera—then mix them together.

3 classes
🔒Private by design: uploaded and camera image files stay in this browser. When you sign in, Project Cloud can sync the project structure and AIM sample references, but it does not upload your private image bytes.
2

LOOK AT YOUR DATA

Is your training set ready?

A good learner needs enough different examples of each thing.

Needs examples

Class balance

The data-quality check compares how many examples each class has.

Data quality coach

TRAINING0images the model learns from
VALIDATION—kept aside to check learning
UNSEEN TEST—never used during training
DIVERSITY READINESS0/100
estimate from quantity, balance and source variety
Build a dataset to see how the experiment separates training, validation and unseen testing.
3

TRAIN

Let the machine learn patterns

The computer vision model converts image pixels into numerical visual patterns and learns which class those patterns usually belong to.

Not trained

Vision engine preparing…

You can add training images while AIM prepares the browser vision engine.

Preparing local vision

DEEP VISION CONTROLS · enable the optional engine to use these

4

TEST

Test the model with something new

Use an image the model did not train on. The model will compare it with every learned class and show its scores.

Waiting for model
◎ Choose a new test image

Do not reuse one of the exact training pictures if you want a fair test.

◌No prediction yet

Train your model, then test it with a new image.

CHANGE + UNDERSTAND

Now experiment like a scientist

Change only one thing: add more examples, vary the background, remove weak examples or change a class. Retrain and compare what happens.