Computer Vision Laboratory

This repository contains a collection of computer vision experiments and a mini project focusing on different image processing and computer vision techniques.

Experiments

1. Edge Detection - Implementation of various edge detection techniques
2. Difference of Gaussians (DoG) - Image enhancement using DoG
3. Histogram of Oriented Gradients (HoG) - Feature extraction using HoG
4. Harris Corner Detector - Corner detection in images
5. SIFT - Scale-Invariant Feature Transform implementation
6. ORB Descriptor - Oriented FAST and Rotated BRIEF features
7. Kmeans - Image segmentation using Kmeans clustering
8. Gabor Filter - Texture analysis using Gabor filters
9. Gaussian Mixture Model - Probabilistic modeling

Mini Project: 3D Model Generation from Images using MoGe

The mini project demonstrates how to generate 3D models from single images using Microsoft's MoGe (Monocular 3D Generation) model.

Dependencies

• PyTorch
• OpenCV
• Trimesh
• NumPy
• PIL
• MoGe model

How it Works

1. The code clones the Microsoft MoGe repository
2. Installs all required dependencies
3. Loads a pre-trained MoGe model from Hugging Face
4. Processes an input image to generate 3D coordinates, depth maps, and normals
5. Creates a textured 3D mesh using the extracted information
6. Exports the 3D model in PLY format

Sample Usage

# Load the model from huggingface hub
model = MoGeModel.from_pretrained("Ruicheng/moge-vitl").to(device).eval()

# Read the input image
input_image = cv2.cvtColor(cv2.imread("input.png"), cv2.COLOR_BGR2RGB)

# Convert image to tensor
input_image_tensor = torch.tensor(input_image / 255, dtype=torch.float32, device=device).permute(2, 0, 1)

# Infer 3D information
output = model.infer(input_image_tensor, resolution_level=9, apply_mask=True)

# Generate and export mesh
mesh.export('output_model.ply')

Instructions

Each experiment is contained within its own directory with a Jupyter notebook (.ipynb) file that includes detailed instructions, code implementation, and explanations.

To run the experiments:

1. Open the corresponding notebook in Jupyter
2. Execute cells sequentially
3. Follow the instructions within each notebook

For the mini project, a GPU environment is recommended for faster processing.