Part 1:
Concept: Train an AI to recognize toy vehicles and various hot wheels cars and model cars. I wanted to train an AI to recognize several Hot Wheels cars and differentiate them.
Train AI to recognize several vehicles based on models on a neutral background
Part 2:
Also experimentation with various AI image generation and how they render vehicles differently.
Images trained on neutral background to mitigate AI confusion with different environments
Tools
The tools I used for this project include Google's Teachable Machine. I used this to train the AI to differentiate objects.
Ml5 was used to obtain the image classifier class to incorporate the teachable machine.
P5JS was used as the editor to edit, code and compile.
Demo
The AI recognizes the trained model and notes on the screen the image label if a certain threshold of confidence is passed.
Part 2.
AI Image Generation Experimentation
AI Generator 1 - "night cafe"
AI Generator 2 - Google Imagen Research Tool
These tools are similar, however Google's Imagen is much more advanced with much more realsistic images that are seamlessly put together.
Compare the examples from Google's Imagen (Top) vs Night cafe's interpretation of my rather simple input of 2 vehicle scenarios.
Another concept I thought about and researched during this project was bias in image generation, specifically the color of one's hand may alter the results.
Additionally, on a larger scale, bias in more complex imagery regarding different types of people, locations, cultures and environments.
A beta version of the Google Imagen, based on DALL-E is available on the above link.
Thank you!
