I've been wanting to dive into the world of AI by building a small project on my own. I want to get my hands dirty coding, training, and experimenting instead of just relying on APIs. For those who have experience with this, I'd love to know: do you think it's better to start with a framework like PyTorch or TensorFlow, or is there a higher-level option that's friendlier for beginners? Also, how small can my model realistically be while still producing interesting results? And do you have any tips for managing datasets and preprocessing without feeling overwhelmed?
2 Answers
When you say 'from scratch,' do you mean collecting your own data? That's a fantastic way to start! Just keep in mind that for tasks like classifying images, you'll generally need a lot of data—like millions of examples for complex images. If you simplify your problem, like classifying books based on their dimensions, you might only need around 100 samples to start seeing results.
Check out this great series of tutorials! They were live-streamed, so you get to see everything in real-time—perfect for understanding the whole process step by step.
Actually, you don’t need millions for a male/female classification model. There's research showing that with consistent data, you can do well with just a few thousand images, or even 50 for specific cases. It's all about the signal-to-noise ratio!