Gastrointestinal disease detection is an important task in medical image analysis, and the Hyper-Kvasir dataset is a widely-used benchmark dataset for this task. ResNet (Residual Network) is a deep learning architecture that has achieved state�of-the-art performance on many image classification tasks. This report presents an implementation of ResNet50, a deep learning model, for the detection of gastrointestinal diseases using the hyper-kvasir dataset. The dataset contains 8,000 high-resolution images of the gastrointestinal tract, labeled with 23 different classes, and we have selected four classes for this project. We performed data preprocessing, including resizing images and data augmentation, followed by fine-tuning the pre-trained ResNet50 model and training the model using the Adam optimizer. The model's performance was evaluated on the testing set, and the results showed an accuracy of around 92% after five epochs. Overall, the implementation of ResNet50 on the hyper-kvasir dataset shows promising results in detecting gastrointestinal diseases, and further improvements can be made by optimizing the hyperparameters and experimenting with other deep-learning architectures.
The Code folder contains the google colab notebooks.
The Dataset contains the image dataset and video dataset used in the codes. The dataset are taken from the Hyper-kvasir dataset. Hyper-Kvasir dataset link: https://datasets.simula.no/hyper-kvasir/
The Reports folser contains the mid-semester and end-semester reports and the weekly reports.
The presentation folder contains the mide-semester and end-semester presentations.
The Results folder contains the images of the results of different models.
Group 8: Dev Patel AU2040056, Meet Patel AU2040010, Manthan Patel AU2040123, Sagar Bajaj AU2040252