Hear from one of the teams working on ICDSS’s Satellite Inpainting Advanced Analytics Project 2021/2022
Project Summary
This research project involves inpainting satellite images using techniques called neural processes of which there are several variants. None of us had learned about this, so our first task was to carry out a literature review of the area. This included reading about meta learning and neural processes through papers, websites and implementations (check out https://yanndubs.github.io/Neural-Process-Family/text/Intro.html#). We also needed to understand how we were going to process the satellite images to be inpainted. The data was collected using Google Earth Engine and a useful script by Harrison Zhu (our PhD mentor). This was our main focus in Autumn term while many of us had exams and coursework.
This term, our aim has been on getting some results based on a variant of neural processes called Convolutional Conditional Neural Processes (ConvCNPs). Luckily, this has already been implemented in a library so some of us are working on understanding this and getting it to run with our satellite images. The rest of us are working on collecting and processing the data. We expect to get some baseline results soon and then we’ll be working on improving those!
The Team Behind
Alexander Pondaven
Hey, I’m Alex. I’m in my 3rd year of an MEng in Electronic and Information Engineering, and I have a keen interest in deep learning. I had previously worked on video classification projects, so I was attracted to the problem of image inpainting. I also wanted to learn new techniques in machine learning like meta-learning and neural processes. The opportunity to work on a problem that will have an impact on research in the field is very exciting. I hope to get more experience working on research projects and be able to get decent results that can be shared with other researchers, possibly in a paper!
Aixin Zhang
Hi there! My name is Aixin (Alyson) Zhang, and I am a third-year MEng student doing Electrical and Electronic Engineering. With a great interest in machine learning and digital signal processing, I am always keen to learn new concepts and take up exciting projects to improve my problem-solving skills and broaden my horizons. This satellite image inpainting project focuses on image processing using the meta-learning methodology. It stimulates my interest to explore this area. My expectation of this project is to learn how to structure and implement a deep learning project and learn more data processing methods. Also, I am excited to meet and collaborate with other people that share similar interests 🙂
Guo Donghu
Hey everyone! I am Donghu Guo, currently studying MSc Environmental Data Science and Machine Learning programme. I have a strong interest in data analysis and model building and have conducted several Machine Learning projects before. This project can further enrich my experience in deep learning. In addition, I have studied the knowledge of remote sensing in my undergraduate course, which can also be well combined with the project. In this data science society, I met many like-minded partners. I hope to walk with them and explore together 🙂
Hamzah Hashim
Hi, I’m Hamzah, and I am currently a third year student studying Theoretical Physics. I’ve done a couple of small Machine Learning and AI projects before, but this is the first time I have had the opportunity to work on topics at the forefront of the field, such as meta-learning. This project has given me a real feel on what research is like in the world of academia, and has broadened and improved my knowledge and skills in Machine Learning and AI. I hope we are able to get meaningful and useful results out of our work, and hopefully be able to produce a paper from it. 🙂