COMP4423 Computer Vision Assignment 2

This repository contains the implementation and report for COMP4423 Computer Vision Assignment 2.
The project explores CNN architectures (ResNet vs VGG, Deformable CNNs) and 3D scene reconstruction with Tiny NeRF, including ray sampling and volume rendering.

📅 Project Info

• Developed: March 2025

---

📂 Project Structure

src/
│── dCNN.py/               # Implementation of Deformable CNNs
│── Raysampling.py/        # Ray sampling (uniform + stratified) for NeRF
│── ResNetvsVGG.py/        # Comparison of ResNet18 and VGG16 (gradient flow + loss landscape)
│── Tiny_nerf.ipynb/          # Tiny NeRF implementation and experiments
│── Volumne_rendering.py/  # Volume rendering functions and visualization
report.pdf              # Detailed assignment report with results and explanations
COMP4423 Assignment 2 - requirements.pdf        # Assignment specification

---

🚀 Features

• ResNet vs VGG:

  • Gradient flow visualization
  • Loss landscape visualization
  • Comparison and analysis of optimization behavior

• Deformable CNN (dCNN):

  • Custom offset convolution implementation
  • Explanation of channel requirements

• NeRF (Neural Radiance Fields) Components:

  • Ray sampling (uniform & stratified)
  • Volume rendering (opacity, transmittance, weights)
  • Tiny NeRF training & improvement (hidden layer size modification)

• Visualizations:

  • Loss landscapes
  • Gradient flow plots
  • Sampling strategies
  • Volume rendering curves
  • NeRF rendered scenes

---

📖 Report

A full explanation of methods, results, and visualizations can be found in report.pdf.

▶️ Usage

Each part of the assignment is implemented in separate folders under src/.
You can run individual modules as follows:

# Run Deformable CNN
python src/dCNN.py

# Run Ray Sampling
python src/Raysampling.py

# Run ResNet vs VGG comparison
python src/ResNetvsVGG.py

# Run Volume Rendering
python src/Volumne_rendering.py

# Open Tiny_nerf.ipynb in Jupyter Notebook or VS Code and execute the cells step by step:
jupyter notebook src/Tiny_nerf.ipynb

---

📌 Requirements

• Python 3.8+
• PyTorch
• torchvision
• numpy
• matplotlib
• loss-landscapes