Summary of Exploring the world of Computer Vision from the SCBX UNLOCKING AI Seminar (EP4)

Keynote: Exploring the world of Computer Vision

Event: SCBX Unlocking AI EP4, Computer Vision: How AI See Things Like We Do

Collaboration: SCBX and Insiderly.ai

Venue: SCBX NextTech, Siam Paragon, 4th Floor

Speaker: Dr. Samprit Marukthatat Senior Researcher, NECTEC

As technology advances, the power of computers and the so-called Computer Vision has become wider, making it easier for anyone to navigate the vast world.

In the seminar "SCBX UNLOCKING AI: EP4" titled "Computer Vision: How AI See Things Like We Do", Dr. Samprit Maruktat, Senior Researcher, NECTEC, gave a lecture on the topic "Exploring the World of Computer Vision" to explain how this cutting-edge AI technology can help improve people's lives. The key points are as follows:

1. If you want to know what areas Computer Vision can be used for? The easiest way is to watch Hollywood movies such as the 'Iron Curtain' movie The Terminator, where intelligent robots are equipped with all-round high-tech, one of which is to see things and detect them and process them into images, or in the movie Eagle Eyes, where artificial intelligence tries to read people's mouths to see what they are saying.
   

1. One of the things that Computer Vision can make our lives easier right away is to help us find information in various forms, such as searching for photos. Just use many types of generative AI and many platforms that support copyright-free images.
   

What is found in these films? In the past, it may have been seen as unrealistic. But now it is real or likely to become a reality, and not only in these two films it depicts the use of cutting-edge technology through the concept, but there are many other technologies in the movie that have become a reality. However, it may not have been a breakthrough that can be widely used according to the narrative style.

1. Deeper computer vision is being used in many fields, such as medicine, which uses AI to improve image quality to help doctors diagnose X-ray images more accurately, to the use of AI to detect objects, such as detecting tumors in the abdomen. 3D Recognition Modeling, Speech Recognition, etc.
   
2. Dr. Sampharit Explain the principle of computer worldview: Computers see the world as pixels. It looks like a small square that is placed on top of each other to form a large image.
   
3. The first work to use computer vision was design, by writing code to assemble small pixel-level things. Computer Vision began in an era when there was no word AI.
   
4. Neural networks are now being used in computer vision to help automatically extract features from the pixels that make up the image.
   

• In many studies, many research teams have found that the features obtained from these neural networks are more effective than those designed by experts because they are similar and similar.
  
  Because of the combination of each pixel, It is like nerves that come together to form the nervous system.
  
  
  
• Inspiring researchers Creatives have an idea of what they should do if they want to build a big network. This led to the creation of a facial recording system that will record the distinctive features of our faces. It is used to detect moving objects where they are.
• Research on neural networks has been around for a long time, even before AI.
  But in the past, we didn't have a way to build a large network that worked well.
  
  
• We now know what to do to build a large network, and the amount of data and computing power is much better than in the past.
  As a result, the creation and use of large-scale networks, also known as Deep Neural Networks, is widely used.
  
  
• The systems we commonly use are the facial recognition system.
  Or object detection systems such as cars, motorcycles, or people all use Deep Neural Networks.
  
  
• It can be said that all computer vision work today uses deep neural networks.

1. But even though it can be used in many ways today. The challenges of computer vision are not few. If you need to create a large number of images and create a consistent caption in a quick time, how do you control the quality?

Currently, AI can be created that can work on both images and text, such as Stable Diffusion that generates images from captions, Image Captioning systems that generate captions for images, Visual Question Answering systems that can answer questions related to images, and OpenAI's CLIP system that helps to see the consistency between images and captions.

In the medical field, such as lung X-ray analysis, the same principle is applied. The system, that is, the neural network used to create features that are suitable for diagnostic imaging, such as lung X-rays of normal people or COVID-19 patients, or thalassemia analysis from blood slide images, also uses Deep Neural Network.

The features created by these neural networks may not be able to be interpreted directly into words. As a result, general doctors do not accept the diagnosis by these neural networks.


Nowadays, there are a lot of doctors who are doing their own research on AI, and there are significantly more research conferences on this topic at medical research conferences. Compared to 5-6 years ago, it is an important mechanism that has led to more and more acceptance of diagnosis by neural networks.

1. Another challenge that developers face is access to data with limited data, making it impossible for researchers and workers to develop quality work. Because often those who have information are not allowed to use it or continue to develop. As a result, the quality of work is not as good as desired.
2. In addition, there are few AI who are good at the Thai language. Compared to the developed foreign models, this will lead to the basic challenges mentioned above, and many more.