Name		Name	Last commit message	Last commit date
parent directory ..
Dataset		Dataset
Graphs		Graphs
Model		Model
Outputs		Outputs
Preprocessing		Preprocessing
Summary		Summary
README.md		README.md
frcnn_test_vgg.ipynb		frcnn_test_vgg.ipynb
frcnn_train_vgg.ipynb		frcnn_train_vgg.ipynb
labels_test.txt		labels_test.txt
labels_train.txt		labels_train.txt
labels_valid.txt		labels_valid.txt
model_vgg_config.pickle		model_vgg_config.pickle
record_1.csv		record_1.csv

README.md

Faster RCNN for Object Detection on Satellite Images

Pre-requistes

Colab
Keras
Tensorflow
Pandas

Preprocessing

In first step, separate images into train,valid and test directories. The dataset was provided in Pascal voc but was converted to custom format of filepath,x1,y1,x2,y2,class_name. Training requires annotation files in .txt format. For step by step preprocessing please refer to Preprocessing/FRCNN_preprocessing.ipynb

Example dataset provided in Dataset folder

Network Summary

Faster RCNN

Training

Refer to frcnn_train_vgg.ipynb

Testing

Refer to frcnn_test_vgg.ipynb.
Note: training config is used for testing the Model.

Performance Measures

Model	Validation mAP	Test mAP
Faster-RCNN	0.515	0.508

Performance Graphs

Training settings

Pretrained weights can be from keras/application

Image Size = 512

Number of RoIs at once = 4

Data Augumentation: Horizontal flips, Vertical Flips, rotation 90 degree.

RPN optimizer: Adam, Learning rate =1e-5

Classification optimizer: Adam, Learning rate =1e-5

Overall optimizer: SGD

Loss function:Mean Absolute Error (MAE)

Visual Results