Project Overview
The Mangrove Propagule Detection and Counting project aims to automate the detection and quantification of mangrove propagules in images and videos using computer vision techniques. This project employs the Faster R-CNN model, a powerful object detection architecture, trained on a custom dataset to enhance the accuracy and efficiency of propagule identification.

Objectives
To develop an automated system for detecting and counting mangrove propagules in various visual inputs.
To enhance conservation efforts by providing reliable data on mangrove populations.
Methodology
Dataset Preparation:

Compiled a dataset of approximately 500 annotated images of mangrove propagules. The annotations were originally in YOLO format and converted to COCO format to ensure compatibility with the Faster R-CNN model.
Model Selection:

Utilized the Faster R-CNN architecture, known for its high accuracy in object detection tasks.
Training Process:

Registered the custom dataset using Detectron2's register_coco_instances method, allowing the model to access training and validation images along with their corresponding annotations.
Configured training parameters, including batch size, learning rate, and the number of classes (one for mangrove propagules).
Data Augmentation:

Implemented augmentation techniques to enhance model robustness and improve generalization across different conditions.
Evaluation:

Evaluated the model's performance using COCO metrics, assessing metrics such as Average Precision (AP) at various Intersection over Union (IoU) thresholds.
Future Work
Optimize model hyperparameters for improved detection accuracy.
Explore advanced data augmentation techniques to enhance model generalization.
Extend the dataset with additional images to improve model training.
Conclusion
This project represents a significant step towards automating mangrove propagule monitoring, ultimately contributing to the conservation and study of these vital ecosystems.