Problem Description and Pipeline
This set of Following videos will talk about:
- How we handle big problem, with very complex machine learning problem
- How to organize todo list (pipeline). This will benefit for us that work as individuals as well as working as a teams
- Introduce another interesting aspects machine learning that will projected by OCR example, and easily practicable at many other complex machine learning.
Photo OCR = Photo Optical Character Recognition
- This could be the case of daily problem that often benefit
- If we look at the photo, we may want to detect the writing in the photo.
There's benefits in the Photo OCR
- As user, we may want to take a photo and when we search through photos we may want to search by input text (Say: "Lula B's Antique Mall) and the photo will be given to us
- For blind people they could take a photo in front of them, and the apps can tell the blind what's the sign in the photo, so they could listen various sign in front of them
- For cars, it may be read the rules or sign on the road, and make a better understanding about its environment, and furthermore proceed to the destination
- While writing recognition may easily found these days, the Photo OCR, where finding the written in the photos, does not. So it still finding the problem where ML can be benefit from
- So these are the steps for POCR pipeline
- First we may want to find the rectangle of the text in the picture
- Second, based on the rectangle, we want to segment the text to characters
- Finally, based on each segmentation, we want to predict the characters inside it
- Some POCR system is more advanced at correcting the predicted characters, but not discussed in the videos.
- This is often called pipeline in Machine Learning
- So it's important to segment the task in ML problem, because it can impacts the overall results.
Each of the task can be assigned by engineer/s or may be designed by one self.
- In summary the pipeline is important, when designing a very complex ML problems.
- In the next set of videos, will be discussed pipeline to introduce another important aspects of ML