app.py

import gradio as gr
from PIL import Image
import torch
from transformers import CLIPProcessor, CLIPModel

# Load model and processor
model = CLIPModel.from_pretrained("laion/CLIP-ViT-H-14-laion2B-s32B-b79K")
processor = CLIPProcessor.from_pretrained("laion/CLIP-ViT-H-14-laion2B-s32B-b79K")

def calculate_similarity(image, text_prompt, similarity_type):
    # Process inputs
    inputs = processor(images=image, text=text_prompt, return_tensors="pt", padding=True)

    # Forward pass
    outputs = model(**inputs)

    # Normalize and calculate cosine similarity
    image_features = outputs.image_embeds / outputs.image_embeds.norm(dim=-1, keepdim=True)
    text_features = outputs.text_embeds / outputs.text_embeds.norm(dim=-1, keepdim=True)
    cosine_similarity = torch.nn.functional.cosine_similarity(image_features, text_features)

    # Adjusting the similarity score based on the dropdown selection
    if similarity_type == "General Similarity (3x scaled)":
        adjusted_similarity = cosine_similarity.item() * 3 * 100
        result_text = f"According to OpenCLIP, the image and the text prompt have a general similarity of {min(adjusted_similarity, 99.99):.2f}%."
    else:  # Cosine Similarity (raw)
        result_text = f"According to OpenCLIP, the image and the text prompt have a cosine similarity of {cosine_similarity.item() * 100:.2f}%."

    return result_text

# Set up Gradio interface
iface = gr.Interface(
    fn=calculate_similarity,
    inputs=[
        gr.Image(type="pil", label="Upload Image", height=512),
        gr.Textbox(label="Text Prompt"),
        gr.Dropdown(label="Similarity Type", choices=["General Similarity (3x scaled)", "Cosine Similarity (raw)"], value="General Similarity (3x scaled)")
    ],
    outputs=gr.Text(),
    allow_flagging="never",
    title="OpenClip Similarity Calculator",
    description="Upload an image and provide a text prompt to calculate the similarity."
)

# Launch the interface with a public link for sharing online
iface.launch(share=True)