Classify Wrinkles — Can my Jeans tell time (1)?
A moment to introspect — can you tell which of your clothes need immediate attention ? Can you tend to your clothes based on their make up, some of them need that gentle , tender care, some of them need you to be more pushy. Some of them are on their own and need your press maybe once a year.
While we are at that, imagine being a laundry detective, effortlessly deciphering the age of your clothes just by glancing at your shelves. Picture setting up a scanner in a mall, instantly notifying people about the freshness of their attire and how long it can last without pressing.
Far fetched it might seem, there are time-tested image classification models based on convolutional learning. These models mimic the human brain and can accurately classify clothes.
The aim of this exploratory analysis or prep work for training is to leverage these image classification models to classify clothes based on their wear and tear and pressing needs.
Before that a brief look at how CNN based training works and how it can be extended for our use case.
Before that , time to rub our favorite magic lamp again:
ChatGPT - explain CNN to a 10 year old + heavy edits
ChatGPT - change this into a poem + light editsIn the realm of circuits and code, a dance of minds unfold,
A team of detectives assemble to search gold,
Convolutional Neural Network, their name so grand,
Mimics the brain’s brilliance, they understand.
With filters keen, they seek out the signs,
In images they search, where pattern aligns,
Edges and curves, their quest underway,
Unraveling shapes, in a pixel array.
Collaborating like stars in the night,
They share their knowledge, a symphony of light,
Complex shapes they unravel, like a puzzle they solve,
As our brains do with grace, mysteries resolve.
Through examples they learn, wisdom they accrue,
Adept at recognizing, things old and new,
Just as we grow, through life’s grand design,
Experiencing, observing, and wisdom refine.
Powerful tools they wield, diverse tasks they embrace,
Recognizing animals, objects in every place,
And in self-driving cars, their potential so bright,
Advancing AI, to a future of light.
Coming back to our example — Machine Learning Practical: 6 Real-World Applications A 5 star DIY course on ML if there is one.
In particular module 2 — Fashion class classification.
Firstly a brief on the train, test data , features
- Fashion training set consists of 70,000 grayscale images.
- Training set contains 60,000 images, and the testing set has 10,000 images.
- Each image is 28x28 pixels, totaling 784 pixels.
- Images are associated with a label from 10 classes representing different clothing items.
- Pixel-values range from 0 to 255 and indicate the lightness or darkness of each pixel.
These are the current categories in the example:
As you might immediately notice only a subset of the above clothing items is relevant for us: 6 -> shirt, 7-> sneaker and maybe 0-> top.
What would our categories look like for our training ?
Next article:
- deep dive into the learning for the fashion data set
- evaluation and the changes which can be done for our “wrinkle free” use case.
- Code snippets and arch diagrams covering phases in training and evaluation.