image2text.py

import streamlit as st
import numpy as np
from PIL import Image

from utils import load_model


def app(model_name):
    model, processor = load_model(f"koclip/{model_name}")

    st.title("Zero-shot Image Classification")
    st.markdown(
        """
        Some text goes in here.
    """
    )

    query = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])
    captions = st.text_input("사용하실 캡션을 쉼표 단위로 구분해서 적어주세요", value="고양이,강아지,느티나무...")

    if st.button("질문 (Query)"):
        if query is None:
            st.error("Please upload an image query.")
        else:
            image = Image.open(query)
            inputs = processor(text=[""], images=image, return_tensors="jax", padding=True)
            # vec = np.asarray(model.get_image_features(**inputs))
            # ids, dists = index.knnQuery(vec, k=10)
            # result_files = map(lambda id: files[id], ids)
            # result_imgs, result_captions = [], []
            # for file, dist in zip(result_files, dists):
            #     result_imgs.append(plt.imread(os.path.join(images_directory, file)))
            #     result_captions.append("{:s} (유사도: {:.3f})".format(file, 1.0 - dist))