app.py

from turtle import color, onclick
import streamlit as st
from PIL import Image, ImageOps
import glob
import json
import requests
import random
import io

random.seed(10)

if 'show' not in st.session_state:
    st.session_state.show = False

if 'example_idx' not in st.session_state:
    st.session_state.example_idx = 0

st.set_page_config(layout="wide")
st.markdown("**This is a demo of the *ImageCoDe* benchmark. What is the task? You are given a description and you have to pick the image it describes, out of 10 images total.**")
st.markdown("**If you click the Sample button, you will get a new text and images. More details of ImageCoDe can be found in our ACL 2022 paper.**")

col1, col2 = st.columns(2)

prefix = 'https://raw.githubusercontent.com/BennoKrojer/imagecode-val-set/main/image-sets-val/'
set2ids = json.load(open('set2ids.json', 'r'))
descriptions = json.load(open('valid_list.json', 'r'))

#example_idx = int(col1.number_input('Sample an example (description + corresponding images) from the validation set', value=0, min_value=0, max_value=len(descriptions)-1))
if col1.button('Sample a description + 10 images from the validation set'):
    st.session_state.example_idx += 1
#    st.session_state.example_idx = random.randint(0, len(descriptions)-1)

img_set, true_idx, descr = descriptions[st.session_state.example_idx]
true_idx = int(true_idx)
images = [prefix+'/'+img_set+'/'+i for i in set2ids[img_set]]
img_urls = images.copy()
index = int(col2.number_input('Image Index from 0 to 9', value=0, min_value=0, max_value=9))

if col1.button('Toggle to reveal/hide the correct image, try to guess yourself before giving up!'):
    st.session_state.show = not st.session_state.show

col1.markdown(f'**Description for {img_set}**:')
col1.markdown(f'**{descr}**') 

big_img = images[index]
img = Image.open(io.BytesIO(requests.get(images[index], stream=True).content))
img_width, img_height = img.size
smaller = min(img_width, img_height)
images[index]= ImageOps.expand(img,border=smaller//18,fill='blue')

caps = list(range(10))
cap = str(index)

if st.session_state.show:
    caps[true_idx] = f'{true_idx} (TARGET IMAGE)'
    img = Image.open(io.BytesIO(requests.get(img_urls[true_idx], stream=True).content))
    img_width, img_height = img.size
    smaller = min(img_width, img_height)
    images[true_idx] = ImageOps.expand(img,border=smaller//8,fill='green')
    if true_idx == index:
        cap = f'{true_idx} (TARGET IMAGE)'
else:
    caps[true_idx] = f'{true_idx}'
    if true_idx == index:
        cap = f'{true_idx}'

col1.image(big_img, use_column_width=True, caption=cap)
col2.image(images, width=175, caption=caps)
col1.markdown(f'{st.session_state.example_idx}')