This repository contains the implementation of a Mask R-CNN model trained for detecting drones in images. The model is built using the Matterport Mask R-CNN implementation and trained on a custom drone dataset.
The Mask R-CNN architecture consists of a backbone network (e.g., ResNet) followed by two subnetworks: a Region Proposal Network (RPN) and a Mask Head. The RPN proposes candidate object bounding boxes, while the Mask Head refines these boxes and predicts binary masks for each object.
The model was trained on a custom drone dataset consisting of 207 images for training and 70 images for validation. The dataset includes images captured from various angles and distances to ensure robust detection performance.
The model was trained for 40 epochs using the Adam optimizer with a learning rate of 0.001. The training process involved optimizing the following losses:
Download the trained weights from the releases section of this repository.
You can view the github repository
here
