app.py

# -*- coding: utf-8 -*-
"""
Created on Mon Apr  7 13:43:34 2025

@author: camaac
"""

import streamlit as st
import os
import random
import pandas as pd
from PIL import Image, ImageEnhance
import numpy as np
import gspread
from oauth2client.service_account import ServiceAccountCredentials
from skimage.exposure import match_histograms
import time
from streamlit_autorefresh import st_autorefresh

# -------------------------
# Global parameters
# -------------------------
IMAGE_DIR = "images"         # Folder containing images
NUM_PAIRS = 25               # Total number of pairs to be assessed
RESULTS_FILE = "results.csv" # CSV file for saving responses

# -------------------------
# Helper functions
# -------------------------
def load_image_pair(index):
    """
    For a given index (integer), returns the path of the ground truth and the path of AI generated image.
    Files are named with a 5-digit index.
    """
    idx_str = str(index).zfill(5)
    gt_path = os.path.join(IMAGE_DIR, f"{idx_str}.png")
    pred_path = os.path.join(IMAGE_DIR, f"{idx_str}_gen0.png")
    return gt_path, pred_path
        
        

def match_brightness(source_img, target_img):
    source_brightness = np.mean(np.array(source_img))
    target_brightness = np.mean(np.array(target_img))

    if target_brightness == 0:
        factor = 1  # avoid division by zero
    else:
        factor = source_brightness / target_brightness

    enhancer = ImageEnhance.Brightness(target_img)
    adjusted = enhancer.enhance(factor)
    return adjusted

def match_histograms_pil(img_reference, img_to_adjust):
    """
    Layer the histogram of `img_reference` on `img_to_adjust`
    (both images are PIL.Image objects).
    Returns a PIL image with adjusted histogram.
    """
    # Convertir les deux images en tableaux numpy
    ref_array = np.array(img_reference)
    adj_array = np.array(img_to_adjust)

    # Ajuster l'histogramme
    matched = match_histograms(adj_array, ref_array, channel_axis=-1)

    # Reconvertir en image PIL
    matched_img = Image.fromarray(np.uint8(matched))

    return matched_img

# -------------------------
# Navigation via st.session_state
# -------------------------
if "page" not in st.session_state:
    st.session_state.page = "intro"
if "user_name" not in st.session_state:
    st.session_state.user_name = ""
if "current_index" not in st.session_state:
    st.session_state.current_index = 0
if "results" not in st.session_state:
    st.session_state.results = []
if "list_pair" not in st.session_state:
    st.session_state.list_pair = []
if "list_pair_ID" not in st.session_state:
    st.session_state.list_pair_ID = []
if "results_tot" not in st.session_state:
    st.session_state.results_tot = 0
if "submitted" not in st.session_state:
    st.session_state.submitted = False
    
# -------------------------
# Intro page
# -------------------------
if st.session_state.page == "intro":
    st.title("AI Wood Generation Evaluation Study")
    st.markdown(
        """
        **Welcome!**

        In this study, you will be shown pairs of wood surface images.
        One image is a real photograph and the other is generated by AI.
        Your task is to select the image you believe is **real**.
        
        ⌛ *Each image pair will be visible for 10 seconds only, be quick!* ⌛

        Please enter your name below and click **Start Evaluation** to begin.
        """
    )
    name = st.text_input("Enter your name:")
    if st.button("Start Evaluation") and name:
        st.session_state.user_name = name
        st.session_state.page = "evaluation"
        st.rerun()
    
    st.session_state.list_pair_ID = random.sample(range(1, 51), NUM_PAIRS)
    

    for i, index in enumerate(st.session_state.list_pair_ID):
        
        gt_path, pred_path = load_image_pair(index)
        
        pair = [("GT", gt_path), ("Pred", pred_path)]
        random.shuffle(pair)
        st.session_state.list_pair.append(pair)
    
    st.stop()



# -------------------------
# Evaluation page
# -------------------------
if st.session_state.page == "evaluation":
    st.title("AI Wood Generation Evaluation")
    # st.write(f"User: **{st.session_state.user_name}**")
    
    if "start_time" not in st.session_state or st.session_state.page_changed:
        st.session_state.start_time = time.time()
        st.session_state.page_changed = False
    
    
    # If all pairs have been evaluated, display a message and save the results
    if st.session_state.current_index+1 > NUM_PAIRS:

        st.markdown("<h4>How confident were you in your answers?</h4>", unsafe_allow_html=True)
        
        confidence = st.radio(   #st.select_slider
            " ",
            [
                "Not confident at all", 
                "Slightly confident", 
                "Moderately confident", 
                "Very confident", 
                "Extremely confident"
            ],
            index=2,   #value="Moderately confident"
            horizontal=True
        )

        if st.button("Submit"):
            
            #Calculating result
            correct_guess = np.array(st.session_state.results)        
            nb_correct = np.sum(correct_guess)
            st.session_state.results_tot = nb_correct

            st.success(f"Number of correct answers: {nb_correct}/{NUM_PAIRS}")
            st.success("Thank you for completing the evaluation!", icon=":material/park:")

            
            #Save result
            scope = ['https://spreadsheets.google.com/feeds', 'https://www.googleapis.com/auth/drive']
            
            creds = ServiceAccountCredentials.from_json_keyfile_name('glass-flux-456209-d4-6fc4b7d9d274.json', scope)
            client = gspread.authorize(creds)
             
            sh = client.open('Results_woodAI').worksheet('test')  

            row = [
                st.session_state.user_name,
                confidence,
                str(nb_correct),
                ",".join(map(str, st.session_state.list_pair_ID)),
                ",".join(map(str, correct_guess)),
            ]
            sh.append_row(row)
            st.session_state.submitted = True
            
            
            # if st.button("See detailed results"):
            #     st.session_state.page = "detailed_results"
            #     st.rerun()
            
        if st.session_state.submitted == True:
            if st.button("See detailed results"):
                st.session_state.page = "detailed_results"
                st.rerun()
            # st.stop()
            
        st.stop()
        
        
    st_autorefresh(interval=1000, key=f"timer_{st.session_state.current_index}")
    st.write(f"Image Pair {st.session_state.current_index+1} of {NUM_PAIRS}")
    
    # Charger et mélanger la paire pour l'index courant
    pair = st.session_state.list_pair[st.session_state.current_index]
    
    img1 = Image.open(pair[0][1])
    img1 = img1.convert("L")
    img2 = Image.open(pair[1][1])
    img2 = img2.convert("L")
    
    # if pair[0][0] == "GT":
    #     img2 = match_brightness(img1, img2)
    # else:
    #     img1 = match_brightness(img2, img1)
    
    elapsed = time.time() - st.session_state.start_time
    
    remaining = max(0, 10 - int(elapsed))
    st.markdown(f"**Time remaining:** {remaining} seconds")
    percent = int((remaining / 10) * 100)
    st.progress(percent)
    
    placeholder = Image.new("L", img1.size, 128)
    
    col1, col2 = st.columns(2)
    
    with col1:
        if elapsed < 10:
            st.image(img1, caption="Image 1", use_container_width=True)
        else:
            st.image(placeholder, caption="Time’s up!", use_container_width=True)
    with col2:
        if elapsed < 10:
            st.image(img2, caption="Image 2", use_container_width=True)
        else:
            st.image(placeholder, caption="Time’s up!", use_container_width=True)
    
    choice = st.radio("Select the real image: ", options=["1", "2"], horizontal = True) #, index=None

    
    if st.button("Next"):
        
        if (choice == "1" and pair[0][0] == "GT") or (choice == "2" and pair[1][0] == "GT"):
            correct_guess = 1
        else:
            correct_guess = 0
        
        st.session_state.results.append(correct_guess)
        
        # Passer à la paire suivante
        st.session_state.current_index += 1
        st.session_state.start_time = time.time()
        st.session_state.page_changed = True
        st.rerun()
        

if st.session_state.page == "detailed_results":
    st.title("Detailed Results")
    results = np.array(st.session_state.results) 
    
    for i in range(NUM_PAIRS):
        pair = st.session_state.list_pair[i]
        
        if pair[0][0] == "GT":        
            imgGT = Image.open(pair[0][1])
            imgGT = imgGT.convert("L")
            imgPred = Image.open(pair[1][1])
            imgPred = imgPred.convert("L")
        else:       
            imgGT = Image.open(pair[1][1])
            imgGT = imgGT.convert("L")
            imgPred = Image.open(pair[0][1])
            imgPred = imgPred.convert("L")
        
        col1, col2 = st.columns(2)
        with col1:
            st.image(imgGT, caption="Real", use_container_width=True)
        with col2:
            st.image(imgPred, caption="AI", use_container_width=True)
        
        result = results[i]
        
        if result:
            st.success("✅ Correct")
        else:
            st.error("❌ Incorrect")
        st.markdown("---")