Technical Column
Author: lyl
Compiled by: Rabbit
What should the new technical column write about? This question has troubled our engineers for a long time. Regarding deep learning, there is an abundance of materials and literature available online; as long as everyone is willing to learn, there is everything from beginner to advanced. Until one day, a friend in our AI technology group suggested sharing some practical cases so that interested and available friends could learn and communicate through practice.
This is indeed a good direction. Knowledge gained from books is always shallow; true understanding requires hands-on experience. There is too much and too scattered theoretical knowledge, and it is difficult to internalize and refine it without practice. Moreover, there are no so-called experts in the world; those who have stumbled often become proficient.This column will engage in practical projects, starting from specific cases, and we will discuss and share experiences in practice together.
Column Outline
In the early stages of the column, we will present three practical projects in the field of computer vision. Below is a brief introduction to each project.
Understanding Neural Networks Through Visualization Techniques
We will introduce how to intuitively understand how convolutional neural networks work using visualization methods through GradCam:
Thought: How does the convolutional neural network complete its tasks?
Directory Detection Based on Lightweight Networks
-
Introduction to the basic principles of SSD
-
Introduction to lightweight models
-
Implementing a general object detector
-
Improvement plans and experiments for face detection based on SSD
GAN Techniques and Their Interesting Applications
-
Basic principles of GAN
-
CycleGAN for image transformation between two domains
-
StarGAN for image transformation between multiple domains
-
Font style transfer
Of course, we will also update some other practical projects later, and we welcome everyone to propose areas of interest.
Column Features
Valuable experiences and thought collisions.During the practice, we will definitely encounter some problems and reflections, which are what the author most wants to share with everyone through this column. This column will share valuable experiences encountered in the project practice, as well as how to overcome challenges, and we will also collect suggestions and thoughts after each project. For example, whether there are better methods for model optimization, or whether other methods can achieve the same effect, or whether the same methods can be used to do more interesting things, etc. Perhaps only through such exchanges can we make more progress.
Preparation for Practice
If anyone is interested in practicing together, the following preparations are needed:
1. Familiarity with Python programming;
2. Familiarity with the deep learning framework PyTorch; we will mainly conduct experiments based on PyTorch for these projects.
If you have relevant suggestions and ideas, feel free to join our technical communication group for discussions!
