🎯 In this lesson, you will:
- Explore a comprehensive dataset
- See image classification in action
- Start building your own image dataset for classification
📂 Resources
Coolest selfie ever?
Starter - what should a data set look like?
Explore the nasa data set
googlearts NASA's Visual Universe
Key Term 1: Web Scraping
Key Term 2: API
Key Term 3: Clustering
how was this built - web scraped nasas public image archive
used clustering to categorise the images in the impressive word cloud! over 100,00 images!
Project Brief: Teachable Machine
over the next 2 lessons you will be building an image classification model.
Demo: Hand Says Moo
adapatable individual or class depending on webcam availability, have a go at doing image classification to get the hand to prompt different outcomes
Theory: Data Collection
why collect data, whats it got to do with machine learning. data quality - resolution, clarity, how busy the image is etc…
key term true positive
key term true negative
🏗️ Task: Start building your own data set
for an image classifier, make 2 folders one called positives, one called negatives
folder one is going to be examples of the main thing you want to identify (true postives)
folder two is going to negative examples, images that are not the thing you want to identify (true negatives)
some examples if you are stuck:
reading facial expressions - is someone smiling?
identifying different dog breeds
identifying road traffic - cars or pedestrians
IMPORTANT all the images must be of exactly the same category (e.g. just dogs, not dogs, cars, bikes or anything else you are interested in, the focus of the task is to have distinct category types)
Please check the example datasets below before starting!
extension open teachable machine and start training your model, must have collected at least 20 images for positives and negatives first!!