COCO Results Analysis

This repo contains a simple toolkit to analyze object detection results of MS-COCO style JSONs.

Description

Given COCO style groundtruth and prediction JSONs, you can analyze the per class Average Precision and Average Recall, and also plot the confusion matrix!

Getting Started

Dependencies

Install all the packages needed through pip:

pip install -r requirements.txt

Executing program

Run the following command:

python main.py --gt_path path_to_gt_json --pred_path path_to_pred_json --save_dir path_to_save_output

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

ConfusionMatrix class is adapted from:

• object_detection_confusion_matrix

Per class Average Precision and Average Recall calculation are adapted from:

• YOLOX