COS 429 Final Project: Interpretability in Pedestrian Detection

Authors: Indu Panigrahi and Raymond Liu

Included in this repository:

• detectron2 zip file - Contains custom Detectron2 module and custom pycocotools module adapted for this project.
• detectron2_fasterrcnn.ipynb - Jupyter notebook for using Detectron2 Faster R-CNN with a subset of the Caltech Pedestrian Dataset
• caltech_subset - Contains json annotations files in the COCO format of the training, validation, and test sets used for this project.
• Poster presented at the COS 429 Poster Session on 12/13/2021
• Submitted Project Report

Note: The Caltech Pedestrian Dataset is formatted as a series of .vbb files. To use them with Detectron2, we first converted the dataset into the PASCAL VOC format using this repository and then converted the resulting dataset to the COCO format using this repository.