Read on Twitter
Do you wanna start getting your hands dirty with Machine Learning and Computer Vision?
Here you have 10 projects to start practicing and improve your portfolio.
Here you have 10 projects to start practicing and improve your portfolio.
If you start any of these projects, try and take them beyond building a Computer Vision model and spend some time making them an end-to-end application.
This is a great way to get experience in building fully-fledged solutions that solve a problem from top to bottom.
This is a great way to get experience in building fully-fledged solutions that solve a problem from top to bottom.
1. Rock, Paper, Scissors
Make a webpage that plays this game using the webcam as the input method.
The computer will generate a random choice and will evaluate the winner by looking at your hand gesture.
This dataset will help: https://www.tensorflow.org/datasets/catalog/rock_paper_scissors
Make a webpage that plays this game using the webcam as the input method.
The computer will generate a random choice and will evaluate the winner by looking at your hand gesture.
This dataset will help: https://www.tensorflow.org/datasets/catalog/rock_paper_scissors
2. Classifying handwritten digits
Given a handwritten address, your application should recognize and display the zip code.
Do not use any OCR libraries. Instead, focus on building a deep learning model to solve this problem.
Can the MNIST dataset help?
Given a handwritten address, your application should recognize and display the zip code.
Do not use any OCR libraries. Instead, focus on building a deep learning model to solve this problem.
Can the MNIST dataset help?
3. Identifying house numbers
Given a street view photo, recognize and display every house number you can find.
It's really fun to go around town and trying this out taking random pictures.
This dataset may help: http://ufldl.stanford.edu/housenumbers/
Given a street view photo, recognize and display every house number you can find.
It's really fun to go around town and trying this out taking random pictures.
This dataset may help: http://ufldl.stanford.edu/housenumbers/
4. Tracking your face
Build a webpage that displays a live view using your webcam. There should be a square around your face, and no matter how you move it, the square should always stay locked in your face.
OpenCV is your friend here. But you can also use Deep Learning.
Build a webpage that displays a live view using your webcam. There should be a square around your face, and no matter how you move it, the square should always stay locked in your face.
OpenCV is your friend here. But you can also use Deep Learning.
5. Photo Sketching
Given an input image, return a sketch or cartoon version of it.
Have fun with this one! You can do all sorts of creative transformations to the input image.
OpenCV will have everything you need for this.
Given an input image, return a sketch or cartoon version of it.
Have fun with this one! You can do all sorts of creative transformations to the input image.
OpenCV will have everything you need for this.
6. Blurring faces
Given a photo, return a version of it with every face blurred.
This is great to hide the identity of people.
Bonus: Can you do the same using a video as an input?
Given a photo, return a version of it with every face blurred.
This is great to hide the identity of people.
Bonus: Can you do the same using a video as an input?
7. Counting people
Given a photo, return the number of people that appear in it. This is the basis to work with algorithms that control crowds.
You could extend this (advanced!) to work on a video feed, and mark people crossing an invisible line in the scene.
Given a photo, return the number of people that appear in it. This is the basis to work with algorithms that control crowds.
You could extend this (advanced!) to work on a video feed, and mark people crossing an invisible line in the scene.
8. Detecting when something is out of place
Given a picture of your desk, determine whether something is out of place compared to how it's normally organized (no need to say which object was moved.)
This one is fun! Hint: Look into autoencoders.
Given a picture of your desk, determine whether something is out of place compared to how it's normally organized (no need to say which object was moved.)
This one is fun! Hint: Look into autoencoders.
9. Classifying traffic signs
Build an application that, given a picture, recognizes any traffic signs in it.
Start with a couple of different signs. You can add more as you go.
Dataset: https://cg.cs.tsinghua.edu.cn/traffic-sign/
Build an application that, given a picture, recognizes any traffic signs in it.
Start with a couple of different signs. You can add more as you go.
Dataset: https://cg.cs.tsinghua.edu.cn/traffic-sign/
10. Style transfer
Given a painting of your favorite artist, and any other random picture, the application should output the same picture using the style from the painting.
This one is not easy, but the results are pretty cool.
Given a painting of your favorite artist, and any other random picture, the application should output the same picture using the style from the painting.
This one is not easy, but the results are pretty cool.
Have you worked on any Computer Vision projects that you'd like to share?
Let me know in the comments below!
Let me know in the comments below!
