In 2019 I did my academic internship on machine learning
What it does
I extracts information from students-id(s) within images.
How I built it
I built the project using object detection, instance segmentation, image classification and image alignment.
Challenges I ran into
I had two main difficulties while developing the project and writing this tutorial: My personal computer has no GPU, so I had to rely on Google Colaboratory to train my models. Also, due to the issue of everything licenced under the MIT licence, I had to create my own datasets for the project.
Accomplishments that I'm proud of
I'm really proud to have developed and implemented a methodology for information extraction from scratch using my best machine learning library, Pytorch. Mostly because Information Extraction is real has been a real challenge in the discipline of computer vision for so many years now.
What I learned
I've mostly learned how to optimize my Pytorch code. I also learned a few tricks for developing Information Extraction Systems.
What's next for Pytorch-Information-Extraction
What comes next for this tutorial is surely the following: Make the detection module more robust to truncated documents. Develop and train a custom OCR engine using Pytorch.