text2image.py

import io
import os
import requests
import zipfile
import natsort
import gc

os.environ["TOKENIZERS_PARALLELISM"] = "false"
from stqdm import stqdm
import streamlit as st
from jax import numpy as jnp
import transformers
from transformers import AutoTokenizer
from torchvision.transforms import Compose, CenterCrop, Normalize, Resize, ToTensor
from torchvision.transforms.functional import InterpolationMode
from modeling_hybrid_clip import FlaxHybridCLIP

import utils


@st.cache(hash_funcs={FlaxHybridCLIP: lambda _: None})
def get_model():
    return FlaxHybridCLIP.from_pretrained("clip-italian/clip-italian")


@st.cache(
    hash_funcs={
        transformers.models.bert.tokenization_bert_fast.BertTokenizerFast: lambda _: None
    }
)
def get_tokenizer():
    return AutoTokenizer.from_pretrained(
        "dbmdz/bert-base-italian-xxl-uncased", cache_dir="./", use_fast=True
    )


@st.cache(suppress_st_warning=True)
def download_images():
    # from sentence_transformers import SentenceTransformer, util
    img_folder = "photos/"
    if not os.path.exists(img_folder) or len(os.listdir(img_folder)) == 0:
        os.makedirs(img_folder, exist_ok=True)

        photo_filename = "unsplash-25k-photos.zip"
        if not os.path.exists(photo_filename):  # Download dataset if does not exist
            print(f"Downloading {photo_filename}...")
            response = requests.get(
                f"http://sbert.net/datasets/{photo_filename}", stream=True
            )
            total_size_in_bytes = int(response.headers.get("content-length", 0))
            block_size = 1024  # 1 Kb
            progress_bar = stqdm(
                total=total_size_in_bytes
            )  # , unit='iB', unit_scale=True
            content = io.BytesIO()
            for data in response.iter_content(block_size):
                progress_bar.update(len(data))
                content.write(data)
            progress_bar.close()
            z = zipfile.ZipFile(content)
            # content.close()
            print("Extracting the dataset...")
            z.extractall(path=img_folder)
    print("Done.")


@st.cache()
def get_image_features(dataset_name):
    if dataset_name == "Unsplash":
        return jnp.load("static/features/features.npy")
    else:
        return jnp.load("static/features/CC_embeddings.npy")


@st.cache()
def load_urls(dataset_name):
    if dataset_name == "CC":
        with open("static/CC_urls.txt") as fp:
            urls = [l.strip() for l in fp.readlines()]
        return urls
    else:
        ValueError(f"{dataset_name} not supported here")


def get_image_transform(image_size):
    return Compose(
        [
            Resize([image_size], interpolation=InterpolationMode.BICUBIC),
            CenterCrop(image_size),
            ToTensor(),
            Normalize(
                (0.48145466, 0.4578275, 0.40821073),
                (0.26862954, 0.26130258, 0.27577711),
            ),
        ]
    )


def app():

    st.title("From Text to Image")
    st.markdown(
        """
    
        ### 👋 Ciao!

        Here you can search for images in the Unsplash 25k Photos dataset and the Conceptual Caption dataset.
        You will see most queries make sense. When you see errors, there might be two possibilities: the model is answering
        in a wrong way or the image you are looking for and the model is giving you the best answer it can get.
        
        
        
        🤌 Italian mode on! 🤌


        You can choose one of our examples down below...
        """
    )

    suggestions = [
        "Un gatto",
        "Due gatti",
        "Un fiore giallo",
        "Un gatto sopra una sedia",
    ]
    sugg_idx = -1

    col1, col2, col3, col4 = st.beta_columns([1, 1, 1, 2])
    with col1:
        if st.button(suggestions[0]):
            sugg_idx = 0
    with col2:
        if st.button(suggestions[1]):
            sugg_idx = 1
    with col3:
        if st.button(suggestions[2]):
            sugg_idx = 2
    with col4:
        if st.button(suggestions[3]):
            sugg_idx = 3

    col1, col2 = st.beta_columns([3, 1])
    with col1:
        query = st.text_input("... or insert an Italian query text")
    with col2:
        dataset_name = st.selectbox("IR dataset", ["Unsplash", "CC"])

    query = suggestions[sugg_idx] if sugg_idx > -1 else query if query else ""

    if query:
        with st.spinner("Computing..."):

            if dataset_name == "Unsplash":
                download_images()

            image_features = get_image_features(dataset_name)
            model = get_model()
            tokenizer = get_tokenizer()

            if dataset_name == "Unsplash":
                image_size = model.config.vision_config.image_size
                dataset = utils.CustomDataSet(
                    "photos/", transform=get_image_transform(image_size)
                )
            elif dataset_name == "CC":
                dataset = load_urls(dataset_name)
            else:
                raise ValueError()

            image_paths = utils.find_image(
                query, model, dataset, tokenizer, image_features, 1, dataset_name
            )

        st.image(image_paths)

        gc.collect()

        sugg_idx = -1