Skip to content

arafathosense/Object-Detection-with-Faster-R-CNN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Object Detection with Faster R-CNN

This project implements Object Detection using Faster R-CNN with ResNet50-FPN, a state-of-the-art deep learning model for real-time object localization and classification. The model is fine-tuned on a custom dataset to accurately detect multiple object classes in images, combining region proposal networks (RPN) with a powerful ResNet-50 Feature Pyramid Network.

🏗️ Model Architecture

  • Backbone: ResNet-50 with Feature Pyramid Network (FPN)
  • Detector Head: Faster R-CNN
  • Framework: PyTorch + Torchvision
  • Loss Function: Classification + Bounding Box Regression
  • Optimization: SGD / Adam with learning rate scheduler

📦 Dataset & Training

  • Dataset: Custom dataset prepared for object detection (COCO-style format)
  • Classes: Multiple object categories (e.g., car, laptop, person, bicycle, etc.)
  • Input Size: 224×224
  • Data Split: 80% training / 20% validation
  • Epochs: 10–15
  • Batch Size: 4
  • Hardware: CPU-compatible, GPU-accelerated optional

Training Pipeline:

from torchvision.models.detection import fasterrcnn_resnet50_fpn

model = fasterrcnn_resnet50_fpn(pretrained=True)
num_classes = len(dataset.classes)
in_features = model.roi_heads.box_predictor.cls_score.in_features
model.roi_heads.box_predictor = FastRCNNPredictor(in_features, num_classes)

📸 Visual Detection Results

image image image

🚧 Future Improvements

  • Convert model to ONNX or TorchScript for deployment
  • Integrate real-time video detection
  • Add custom UI for object annotation
  • Experiment with MobileNet-FPN for faster inference

🙏 Acknowledgements

  • PyTorch & Torchvision team for open-source detection models
  • COCO Dataset for reference annotation format
  • NVIDIA and Kaggle for providing GPU resources

💼 Libraries & Tools

Object Detection using Faster R-CNN (ResNet50-FPN) is powered by a robust deep learning stack — optimized for precision, scalability, and research-ready deployment.

🧠 Every library here plays a vital role — from feature extraction and region proposal to visualization and performance tracking.
🔗 Together, they enable an end-to-end detection pipeline that fuses computer vision and deep learning excellence.

👤 Author

HOSEN ARAFAT

Software Engineer, China

GitHub: https://github.com/arafathosense

Researcher: Artificial Intelligence, Machine Learning, Deep Learning, Computer Vision, Image Processing

About

This notebook performs object detection on manually specified images using a pretrained Faster R-CNN model from torchvision.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors