Abril - Text-Image Comparison using CLIP

Abril is a Python library that makes it easy to compare text and images using OpenAI's CLIP model. It provides a simple interface for calculating similarity scores between text descriptions and images.

Installation

# First, install CLIP:
pip install git+https://github.com/openai/CLIP.git

# Then install abril:
pip install abril

Usage

from abril import ClipComparer
from PIL import Image

# Initialize the comparer
comparer = ClipComparer()

# Compare a single text with an image
score = comparer.compare_text_with_image(
    "a photo of a dog",
    "path/to/image.jpg"
)
print(f"Similarity score: {score}")

# Find the best matching text for an image
texts = [
    "a photo of a dog",
    "a photo of a cat",
    "a photo of a car"
]
best_text, score = comparer.find_best_match(texts, "./image.jpg")
print(f"Best match: {best_text} (score: {score})")

Features

• Easy-to-use interface for text-image comparison
• Support for both single and multiple text comparisons
• Uses CLIP's ViT-B/32 model by default (can be changed)
• Automatically uses GPU if available for faster processing
• Handles both file paths and PIL Images flexibly
• Normalizes features for better comparison
• Returns scores between 0 and 1 for easy interpretation
• Built on OpenAI's CLIP model
• GPU acceleration when available

Requirements

• Python ≥ 3.7
• PyTorch ≥ 1.7.1
• Other dependencies are handled automatically during installation

License

MIT License - see LICENSE file for details.