app.py

import gradio as gr

from . import OffTopicDetector

detector = OffTopicDetector("openai/clip-vit-base-patch32")

def validate(item_id: str, threshold: float):
    images, domain, probas, valid_probas, invalid_probas = detector.predict_item_probas(item_id)
    valid_images = [x for i, x in enumerate(images) if valid_probas[i].squeeze() >= threshold]
    invalid_images = [x for i, x in enumerate(images) if valid_probas[i].squeeze() < threshold]
    return valid_images, invalid_images


with gr.Blocks() as demo:
    with gr.Tabs():
        with gr.Tab("From Item ID"):
            item_id = gr.Textbox(label="Item ID")
            threshold = gr.Number(label="Threshold", value=0.5)
            submit = gr.Button("Submit")
            valid = gr.Gallery(label="Valid images").style(grid=[1, 2, 3], height="auto")
            invalid = gr.Gallery(label="Invalid images").style(grid=[1, 2, 3], height="auto")
            submit.click(inputs=[item_id, threshold], outputs=[valid, invalid], fn=validate)


demo.launch()