Final project for Vision and Cognitive Systems course 2023-2024 held by prof. L. Ballan by Oksana Abramova and Ekaterina Chueva.
This project explores the application of computer vision techniques for skin cancer detection with the task of image classification. The methodology involved data preprocessing, model training and optimisation. We initially constructed a convolutional neural network model (CNN) as a baseline, followed by integrating transfer learning and simple attention mechanism to improve the results. Models of ResNet and DenseNet families were used for transfer learning as well as for exploring the effect of several modifications to the architecture. The results highlight the effectiveness of the chosen architectures in automated skin cancer detection showcasing the potential of these techniques to assist in early diagnosis. The best model in our case was found to be pre-trained ResNet-50 with modifications, which achieved 91.7% accuracy.
The results of the project are shown on the picture below:
Please have a look at the VCS_Abramova_Chueva.pdf file in this repository for the details about the dataset, chosen methodology and considered models.