svdesai / eda-coco

Exploratory Data Analysis and Visualisations on MS COCO style object detection datasets.

Exploratory Data Analysis on MS COCO style datasets.

This repo mainly contains a jupyter notebook in which some basic Exploratory Data Analysis (EDA) is performed on MS COCO. While EDA is widely used for categorical data, I haven't seen a lot of repos doing EDA for object detection datasets, which is the reason I created this repo.

Contents:

• eda_coco_style.ipynb
• coco_train2017.html (EDA results on MS COCO 2017 - train)
• voc2012.html (EDA results on VOC 2012)

Features:

• Some basic high level statistics
• Distribution of objects across images
• Class wise distribution of objects
• Observing average bounding box sizes for each class
• Viewing random images

Notes:

Used this tool to convert PASCAL VOC XML style datasets to MS COCO style JSONs.

Maintainers:

This tool is developed and maintained by Vikas Desai (me).