Which one would you agree with? 😁
“AI: Just a bunch of ones and zeros pretending to be human”
“AI: The future is here, and it’s pretty cool.”
“AI: It’s like having a super smart, slightly sarcastic roommate who never needs a coffee break.”
Before anyone points out, these taglines were suggested by hashtag#Gemini (Generative AI chatbot by Google).
Believe me when I say that everything below, I am just writing as and when some random thoughts are coming in mind after completing this week’s coursework! 😂
Last few days, I have been reading and learning the basics of Artificial Intelligence and the cool world of ‘GenAI’. I find it fascinating to see so many different perspectives on AI and how it can/is changing the lives around us and the way businesses happen. This course I am doing is making me re-think about my experience till now and how advances in AI are solutions to the past problem statements.
This is about Vehicle Inspections! Yes I worked for some time in the used-car industry.
I recall few years back, I was working on building & improving a used-vehicle inspection system and we used to discuss a lot about how cool it would be if we could click pictures of a car and the system could tell if there are scratches on the door surface, or any visible dent along with dimensions of the dent & the impact. And then all this data going into some intelligent system that returns the physical/visible health of the vehicle. There were some early companies who were working on building this but no major breakthrough until few years back. This led me back to my new friend Google Gemini. 😊
Here is what I learnt, on how we will build it today! (Maybe this post can end with a generic solution to the problem statement).
Typically, in general and being absolutely concise, AI solutions go through some standard steps.
First we need some diverse dataset. Possible a very details and varied data set. Second, we design the model architecture Third, we train the model. Fourth, we use it for our intended use-case.
This is exactly how we solve for our use-case inspection problem statement.
- Data Collection and Preparation: Collect hundreds of thousands of car images with different types of dents, issues and different types of cars. Label the dents in the dataset and add augment the data to increase diversity.
- Model Architecture: Extract dent information from the dataset, reduce dimensions, and then eventually categorise the images if they have dents.
- Train the model. Measure the performance. Optimize the model.
- Test the model on a separate test dataset and measure for accuracy, precision and errors.
- And eventually you will have a model to identify dents.
Will seems simple steps but each step is a story in itself. I was thinking I might dwell deeper into each step in my next post.
If you were able to read through the last of this loooong post, congratulations! I hope it was joyful. 🤓
#AI #GenAI
Leave a Reply