4-Credit Project for CS 444 Deep Learning for Computer Vision
This project fine-tunes pre-trained Mask R-CNN and SOLO models on the Penn-Fudan pedestrian dataset and evaluates their instance segmentation performance using COCO metrics. You can see the results of my project in my research paper.
To train and run the models, follow CS 444 Final Project.ipynb.
PyTorch tutorial: https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html#
Penn-Fudan dataset from Kaggle: https://www.kaggle.com/datasets/psvishnu/pennfudan-database-for-pedestrian-detection-zip
SOLO model implementation: https://github.com/zew013/SOLOv2--Dynamic-and-Fast-Instance-Segmentation