Salient Object Detection (SOD) Project

This repository contains an end-to-end Salient Object Detection (SOD) system built using a custom U-Net++ deep-learning model in TensorFlow/Keras.
The objective of the project is to generate pixel-level saliency masks that highlight the most important object in an image.

---

Project Documentation

You can view the full project files here:

• Final Report (PDF)
  Open the report

• Project Presentation (PPTX)
  Open the presentation

---

What This Project Includes

• Project Report — dataset description, preprocessing, model architecture, training setup, evaluation, and results
• Presentation Slides — summarized visual explanation of the project
• Source Code — scripts for:
  • dataset loading & preprocessing
  • U-Net++ model construction
  • training
  • evaluation & visualization

---

Project Summary

• Dataset: ECSSD
• Model: U-Net++ encoder–decoder
• Framework: TensorFlow / Keras
• Task: Pixel-wise saliency prediction
• Metrics: IoU, Precision, Recall, F1 Score

The trained model successfully learned to detect salient objects and produced consistent, accurate saliency masks on unseen test images.

---