app.py

import ripple
import streamlit as stl
from tqdm.auto import tqdm

# streamlit app
stl.set_page_config(
    page_title="Ripple",
)

stl.title("ripple search")
stl.write(
    "An app that uses text input to search for described images, using embeddings of selected image datasets. Uses contrastive learning models(CLIP) and the sentence-transformers"
)
stl.link_button(
    label="Full library code",
    url="https://github.com/kelechi-c/ripple_net",
)

dataset = stl.selectbox(
    "choose huggingface dataset(bigger datasets take more time to embed..)",
    options=[
        "huggan/few-shot-art-painting",
        "huggan/wikiart",
        "zh-plus/tiny-imagenet",
        "huggan/flowers-102-categories",
        "lambdalabs/naruto-blip-captions",
        "detection-datasets/fashionpedia",
        "fantasyfish/laion-art",
        "Chris1/cityscapes"
    ],
)
# initalized global variables

embedded_data = None
embedder = None
finder = None
search_term = None
ret_images = []
scores = []

#@stl.cache_data
def embed_data(dataset):
    embedder = ripple.ImageEmbedder(
            dataset, retrieval_type="text-image", dataset_type="huggingface"
    )
    embedded_data = embedder.create_embeddings(device="cpu")
    return embedded_data

@stl.cache_resource
def init_search(embedded_data):
    text_search = ripple.TextSearch(embedded_data, embedder.embed_model)
    stl.success("Initialized text search class")
    return text_search

def get_images_from_description(description):
    scores, ret_images = finder.get_nearest_examples(description, k_images=4)
    return scores, ret_images
    

if dataset and stl.button("embed image dataset"):
    with stl.spinner("Initializing and creating image embeddings from dataset"):
        embedded_data = embed_data(dataset)
        stl.success("Successfully embedded and created image index")

if embedded_data:
    finder = init_search(embedded_data)
    
    search_term = stl.text_input("Text description/search for image")

if search_term:
    with stl.spinner("retrieving images with description.."):
        scores, ret_images = get_images_from_description(search_term)
        stl.success(f"sucessfully retrieved {len(ret_images)} images")

try:
    for count, score, image in tqdm(zip(range(len(ret_images)), scores, ret_images)):
        stl.image(image["image"][count])
        stl.write(score)
        
except Exception as e:
    st.error(e)