🌍 Disaster Image Classification using Vision Transformer

This project uses a fine-tuned Vision Transformer (ViT) model to classify disaster-related images into various categories such as Water Disaster, Fire Disaster, Human Damage, etc.

🚀 Installation

Install the required Python packages:

pip install transformers torch torchvision pillow requests

🔍 Quick Start

Use the pipeline to classify an image directly from a URL:

from transformers import pipeline
from PIL import Image
import requests
from io import BytesIO

# Load the image classification pipeline
pipe = pipeline("image-classification", model="Luwayy/disaster_images_model")

# Load an image from a URL
url = 'https://www.spml.co.in/Images/blog/wdt&c-152776632.jpg'
response = requests.get(url)
image = Image.open(BytesIO(response.content))

# Classify the image
results = pipe(image)

# Print results
print(results)

Example Output:

[
  {"label": "Water_Disaster", "score": 0.9184},
  {"label": "Land_Disaster", "score": 0.0200},
  {"label": "Non_Damage", "score": 0.0169},
  {"label": "Human_Damage", "score": 0.0164},
  {"label": "Fire_Disaster", "score": 0.0143}
]

🧠 Model Details

Base Model: google/vit-base-patch16-224-in21k
Architecture: Vision Transformer (ViTForImageClassification)
Image Size: 224x224
Classes:
- Damaged_Infrastructure
- Fire_Disaster
- Human_Damage
- Land_Disaster
- Non_Damage
- Water_Disaster

⚙️ Training Configuration

Image Normalization: Mean [0.5, 0.5, 0.5], Std [0.5, 0.5, 0.5]
Resize Method: Bilinear to 224x224
Augmentations: Resize, Normalize, Convert to Tensor
Batch Size: 16
Epochs: 3
Learning Rate: 3e-5
Weight Decay: 0.01
Evaluation Strategy: Per epoch