Repository contains RetinaNet,Yolov3 and Faster RCNN for multi object detection on satellite images dataset.

Further Details can be found here in repsective Readme Files, contains Preprocessing, performance graphs, visual results, network summaries etc.

RetinaNet
Yolov3
Faster RCNN

Satellite Imagery Multi-vehicles Dataset (SIMD). It comprises 5,000 images of resolution 1024 x 768 and collectively contains 45,303 objects in 15 different classes of vehicles including cars, trucks, buses, long vehicles, various types of aircrafts and boats. The source images are taken from public satellite imagery available in Google Earth and contain images of multiple locations from seven countries.

Access Complete Dataset here: http://vision.seecs.edu.pk/simd/

With each image, the annotation is available as text file. The annotation format can be described as (c, xi, yi, w, h), where c is the object class starting from 0, (xi, yi) are the center of object and (w, h) are width and height respectively. All these values are percentages to the actual image.

This repository contains three different object detection model alongwith their improvements:

• Yolov3
• introduced SPP (Spatial Pyramid Pooling) module to Yolov3
• RetinaNet
  • (1) supports ResNet50 backbone
  • (2) supports EfficientNetB7 backbone
• Faster-RCNN
  • supports VGG16

Pre-trained models can be downloaded from */Model/link.txt in respective folder.

Asim Hameed Khan
Contact: https://www.linkedin.com/in/asimniazi63/

https://github.com/RockyXu66/Faster_RCNN_for_Open_Images_Dataset_Keras/
https://github.com/ultralytics/yolov3
https://github.com/fizyr/keras-retinanet