app.py

import torch
import gradio as gr
from transformers import AlignProcessor, AlignModel


device = "cuda" if torch.cuda.is_available() else "cpu"

processor = AlignProcessor.from_pretrained("kakaobrain/align-base")
model = AlignModel.from_pretrained("kakaobrain/align-base").to(device)
model.eval()


def predict(image, labels):
    labels = labels.split(', ')
    inputs = processor(images=image, text=labels, return_tensors="pt").to(device)

    with torch.no_grad():
        outputs = model(**inputs)

    logits_per_image = outputs.logits_per_image  
    probs = logits_per_image.softmax(dim=1).cpu().numpy()
    return {k: float(v) for k, v in zip(labels, probs[0])}


description = """
<div class="container" style="display:flex;">
  <div class="image">
    <img src="https://upload.wikimedia.org/wikipedia/commons/thumb/7/73/Rafale_-_RIAT_2012_%287852045796%29.jpg/1200px-Rafale_-_RIAT_2012_%287852045796%29.jpg" alt="rafale" />
  </div>
  <div class="text">
  <p>Compare anything</p>
  </div>
</div>
"""

gr.Interface(
    fn=predict,
    inputs=[
        gr.inputs.Image(label="Image to classify", type="pil"),
        gr.inputs.Textbox(lines=1, label="Comma separated candidate labels", placeholder="Enter labels separated by ', '",)
    ],
    theme="grass",
    outputs="label",
    examples=[
        ["cartoon.jpeg", "dinosaur, drawing, forest",], 
        ["painting.jpeg", "watercolor painting, oil painting, boats",], 
    ],
    title="Zero-Shot Image Classification with ALIGN",
    description=description
).launch()