Coyote: A Dataset of Challenging Scenarios in Visual Perception for Autonomous Vehicles

The project aims to curate a dataset of naturally-existing images that can potentially obstruct the performance of Autonomous Vehicles. It also provides a demonstration by testing the collected images on the state-of-the-art Object Detection models.

The repository is divided in 5 folders and an Excel file:

/data: The data folder is the heart of the project as it consists of all the images collected for the research.
/models: This folder contains pre-trained models and supporting files for running the inference on state-of-the-art models
/results: The inference results for all the images are stored in this folder
/scripts: This folder consists of all the python files used to run inference for object detection and semantic segmentation tasks.
/report: This folder contains the detailed report about the project submitted as a part of Master
GroundTruth.xlsx: This file contains the ground truth details for all the images in data folder to evaluate the inference results