Since 2018, I change my research focus on person re-identification. With only a little knowledge on Tensorflow and deep learning, it's quite a challenge for me to begin in a new area. And when studying, I found no good tutorial in reid. I have to start everything from scratch (not everything). Thus, for saving the world's most intelligent people's time 🙂, I worte this tutorial, or more precisely, share my experience in reid.
This tutorial starts in the begining of 2018, and my knowledge that time is:
- basic math (linear algebra, discrete mathematical, etc.)
- intermediate programming skills (python, numpy)
- basic idea about deep learning
So if you want to read this tutorial, it's recommended that you have at least equal knowledge in the areas above. Also, it's recommended that you watch Stanford CS231n first. The brilliant lecturers will give a clear, general picture of computer vision (of course you should have some general idea about CV before you dive into a specific area). And it's also recommended to finish its homework, which would help you have a better understanding of deep learning framework (Tensorflow, Pytorch, etc.)
In this tutorial, I would assume you have the ability above so programming language related stuffs wont't be covered. You can get everything you need in StackOverflow. This tutorial contains several parts:
- Link to my blog Introduction to Reid - Train a Part-based Convolutional Baseline (PCB)
- Code could be seen in tutorial 2 - pcb
- The model PCB is from paper Beyond Part Models Person Retrieval with Refined Part Pooling