CS231N Final Project (Spring 2025)

Group: Sophia Longo

Project Title: Improving deep learning-based diagnosis of skin malignancies using the Melanoma Research Alliance Multimodal Image Dataset for AI-based Skin Cancer (MIDAS) diagnosis

Github Link: https://github.com/Bio-Programmer/midas

Note: Code transferred from Colab. All input images and model weight files omitted given Github constraints.

Notebooks

1. starter.ipynb: initializing codebase and importing dataests (ISIC, MIDAS)
2. dev.ipynb: training the EfficientNet-B3 model as our "baseline" on the dev set (1,000 ISIC training images)
3. model_experiments.ipynb: testing out differnt models (EfficientNet-B3, Swin) and architecutres (including ensembles) on the dev set (table of results at end)
4. MIDAS.ipynb: after selecting highest performing model from dev experiments in (3), trained on full ISIC training set, fine-tuned on midas
5. plots.ipynb: code for creating plots, confusion matrices from results logs

Directories

1. All models designed and tested can be found in modeles as separate scripts
2. Preprocessing, loading, benchmarking, and evaluation scripts found in utils
3. Training script in src
4. ISIC 2019 data found in data and MIDAS in data_midas; omitted dataset images for upload
5. For each experiment/training round, a training log with metrics, loss curve, and best model weights are saved within a directory in results

Training scheme for MIDAS.ipynb