README

This repository contains two Jupyter notebooks (q1.ipynb and q2.ipynb) demonstrating deep learning models for image classification and image segmentation tasks.

---

1. q1.ipynb: Vision Transformer for CIFAR-10 Classification

Description: Implements a Vision Transformer (ViT) architecture to classify images from the CIFAR-10 dataset into 10 classes.

Key Features:

• Data loading and preprocessing using torchvision.datasets and transforms.
• ViT components:
  • PatchEmbedding to split and embed image patches.
  • Multi-head TransformerEncoderLayer for sequence modeling.
  • Fully connected MLPHead for final classification.
• Training loop with configurable hyperparameters (learning rate, epochs, batch size).
• Evaluation on validation set and accuracy plotting.
• GPU acceleration support via PyTorch.

Dependencies:

• Python 3.8+
• torch
• torchvision
• matplotlib
• numpy

Usage:

1. Install dependencies: pip install torch torchvision matplotlib numpy
2. Run all cells in q1.ipynb.
3. Modify hyperparameters (learning rate, epochs, batch size) in the configuration cell as needed.

---

2. q2.ipynb: Text-Guided Segmentation with SAM and GroundingDINO

Description: Performs zero-shot, text-prompted object segmentation on a sample image using:

• GroundingDINO for region proposal based on text prompts.
• Segment Anything Model (SAM) v1/v2 for precise segmentation masks.

Key Features:

• Installs required libraries: segment-anything, transformers, datasets, accelerate, opencv-python, Pillow, matplotlib.
• Loads an image from a remote URL with requests and Pillow.
• Accepts a text prompt (e.g., "a dog") to specify the target object.
• Uses GroundingDINO to generate bounding box region seeds for the prompt.
• Feeds the region seeds to SAM predictor to produce a binary segmentation mask.
• Displays original image with overlayed segmentation mask.

Dependencies:

• Python 3.8+
• segment-anything
• transformers
• datasets
• accelerate
• opencv-python
• Pillow
• matplotlib
• torch

Usage:

1. Install dependencies:

   pip install segment-anything transformers datasets accelerate opencv-python pillow matplotlib torch

2. Open q2.ipynb and run all cells in order.
3. Input a custom text prompt in the designated prompt cell.
4. Review the generated segmentation mask overlay.