Mafat Challenge - Detection and fine grained classification of objects in aerial imagery

This repo is a pilot (pre-release) implementation of the Computer Vision Mafat Challenge (competition site).
The main differences I'm aware of between this version and the official released-to-public version:

This version	Official version
Detection & Classification	Classification
Classes of Large Vehicle, Small Vehicle, Solar Panel and Utility Pole	Classes of Large Vehicle and Small Vehicle
9369 images	4216 images

Hence, solving this version is different.

Index

• Detecting And Classifying Objects In Aerial Imagery/ folder contains challenge's data (CSVs) and used to contain the dataset as well.
• src/ folder contains code for training & predicting the F-RCNN, ResNet-50 networks as well as several more utility scripts.
• Pilot challenge PDF
• Final report PDF