From data scraping to model optimization: Fedya Chursin's computer vision project

Can you describe the project that you have been working on? 

Fedya: 'The project I've been working on is part of the year 1 curriculum for the ADS&AI programme. It focused on computer vision and spanned over an 8-week period. Each week, we had small assignments that built up to the final project.’  

‘The main idea behind my project was to use computer vision to analyse images of meals and calculate the carbon footprint of the meal. Additionally, the system will analyse eating habits and provide tips to improve personal lifestyle choices with a focus on reducing the carbon footprint.’  

What did developing this tool look like? 

Fedya: ‘We started with a kick-off lecture and then proceeded to work on data preparation, modelling, evaluating deployment, and other steps of the Machine Learning project life cycle. We had self-study days on Monday, Tuesday and Thursday and dedicated time in the data lab on Wednesdays and Fridays to work on the project. I worked on this project alone, taking on the roles of both a Machine Learning engineer and a data scientist.’  

‘To gather images for my data set, I initially used a teacher-provided data set. However, I also created a Python script to scrape the internet for images using the DuckDuckGo API. This way, I ended up with around 5,000 to 7,500 images of different meals. After obtaining the data set, I had to clean it and then I started working on neural networks to create a model that could identify the food items in an image. After that, I trained the model to recognize patterns, such as the shape and colour in food items like beef.’ 

‘Unfortunately, the initial accuracy of the model was only 25%, so I had to improve both the data set and the model itself. I reached out to one of the lecturers for assistance, and they helped improve the data set. By modifying the Python script, performing better preprocessing, and optimizing the model architecture, I was able to achieve an accuracy of 85% to 90%.’  

‘The final step was to make an application. I implemented the machine learning model into the app and spent the last week to design the best user interface and user experience. I created several designs and received feedback from lecturers and peers. Ultimately, I presented my project, including the business idea, marketing research, target audience, and the reasons behind my motivation.’ 

‘During this project, we not only focused on hard skills but also emphasized soft skills such as ethics, fairness, and avoiding biases. It was important for me to consider these aspects throughout the project.’ 

What are your future plans? 

Fedya: ‘I would like to pursue a career as an AI engineer, to develop new models and algorithms. After completing my bachelor's degree, I aspire to pursue a master's degree, and perhaps even a PhD, with a focus on AI applications in the field of medicine.’ 

 

If you would like to know more about Fedya, visit his LinkedIn profile, his account on GitHub and Fiverr, or find his app in the app store.