Course: EE‑655 – Computer Vision & Deep Learning
This repository presents the implementation of a YOLO-based object detection system adapted for complex-valued input. The project is a part of the EE‑655 course curriculum and explores how complex data representations can influence model performance compared to traditional real-valued inputs.
Complex YOLO Implementation.ipynb– Core notebook implementing complex-input YOLO model.Implementation_Demo.ipynb– Notebook demonstrating inference results using trained models.Course_Project.ipynb– End-to-end pipeline integrating data preprocessing, training, and evaluation.LICENSE– MIT License for open-source use.README.md– Project description and usage guide (this file).
git clone https://github.com/naman065/EE-655-Course-Project.git
cd EE-655-Course-ProjectMake sure Python 3.7+ is installed. Then install the necessary libraries:
pip install torch torchvision numpy matplotlib opencv-pythonOther common tools like scikit-learn or tqdm may also be required depending on notebook usage.
Open JupyterLab or Jupyter Notebook and run:
Complex YOLO Implementation.ipynbto train and evaluate the complex-valued model.Course_Project.ipynbto follow the complete workflow.Implementation_Demo.ipynbto visualize model inference.
-
Data Preprocessing
- Load and format dataset.
- Generate both real and complex input representations.
-
Model Training
- Implement YOLO architecture.
- Train on complex-valued and real-valued input data.
-
Evaluation
- Assess model accuracy using standard metrics (e.g., mAP, IoU).
- Compare performance of complex vs. real input pipelines.
-
Inference & Visualization
- Visualize detection outputs on test images.
- Observe qualitative and quantitative differences.
- EE‑655 Lecture Slides & Notes
- YOLOv5 Documentation – https://github.com/ultralytics/yolov5
- Research papers on complex-valued neural networks and object detection
| Name | Roll No |
|---|---|
| Ayush Badgujar | 230259 |
| Ashish Kumar Jha | 230227 |
| Naman Mohan Singh | 230678 |
| Prasun Shrivastav | 230778 |