app.py

import hashlib
import os
import streamlit as st
import requests
from PIL import Image
from io import BytesIO

CACHE_DIR = os.path.join(os.path.dirname(__file__), "image_cache")
os.makedirs(CACHE_DIR, exist_ok=True)


def _cache_path(url: str) -> str:
    """Return the local file path for a cached image URL."""
    key = hashlib.md5(url.encode()).hexdigest()
    return os.path.join(CACHE_DIR, f"{key}.png")


@st.cache_data(show_spinner=False)
def resolve_image_url(url: str) -> str:
    """If the URL is a Wikimedia Commons File: page, resolve it to the direct
    image URL via the MediaWiki API. Otherwise return the URL unchanged.
    Cached in-memory so each File: page is only looked up once per session."""
    if "commons.wikimedia.org/wiki/File:" in url:
        filename = url.split("/wiki/File:")[-1]
        api_url = (
            "https://commons.wikimedia.org/w/api.php"
            f"?action=query&titles=File:{filename}"
            "&prop=imageinfo&iiprop=url&format=json"
        )
        headers = {"User-Agent": "Mozilla/5.0 (compatible; StreamlitApp/1.0; +https://streamlit.io)"}
        try:
            r = requests.get(api_url, headers=headers, timeout=10)
            pages = r.json()["query"]["pages"]
            page = next(iter(pages.values()))
            return page["imageinfo"][0]["url"]
        except Exception:
            return url
    return url


def load_image(url: str):
    """Load an image from disk cache if available, otherwise download it,
    save it to the cache folder, and return a PIL Image.
    The cache persists across app restarts — images are only downloaded once."""
    path = _cache_path(url)
    if os.path.exists(path):
        try:
            return Image.open(path)
        except Exception:
            pass  # corrupted cache file — re-download below

    headers = {
        "User-Agent": (
            "Mozilla/5.0 (compatible; StreamlitApp/1.0; "
            "+https://streamlit.io)"
        )
    }
    try:
        direct_url = resolve_image_url(url)
        response = requests.get(direct_url, headers=headers, timeout=10)
        response.raise_for_status()
        img = Image.open(BytesIO(response.content))
        img.save(path, format="PNG")  # persist to disk
        return img
    except Exception:
        return None


# List of 8 scenarios based on the spreadsheet rules
# Images sourced from Wikimedia Commons (public domain / CC licensed)
# Supports both commons.wikimedia.org/wiki/File: page URLs and direct upload URLs
scenarios = [
    {
        # Scenario 1: Different pictures of the same person (Messi)
        "label": "Different pictures of the same person",
        "image1_url": "https://commons.wikimedia.org/wiki/File:Lionel-Messi-Argentina-2022-FIFA-World-Cup_(cropped).jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Lionel_Messi_in_2018.jpg",
        "answer": "Yes",
        "feedback": "Different photos of the same subject should be marked Yes."
    },
    {
        # Scenario 2: Identical or resized picture (same image, two sizes)
        "label": "Identical or resized picture",
        "image1_url": "https://commons.wikimedia.org/wiki/File:Kevin_Garnett_2008-01-13.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Kevin_Garnett_2008-01-13.jpg",
        "answer": "Yes",
        "feedback": "Identical images (even at different sizes) require a Yes response."
    },
    {
        # Scenario 3: Different pictures of the same landmark (Eiffel Tower)
        "label": "Different pictures of the same landmark",
        "image1_url": "https://commons.wikimedia.org/wiki/File:Tour_Eiffel_Wikimedia_Commons.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Tour_eiffel_at_sunrise_from_the_trocadero.jpg",
        "answer": "Yes",
        "feedback": "Different photos of the same landmark are the same subject."
    },
    {
        # Scenario 4: Subject vs. representation (Eiffel Tower vs. keychain)
        "label": "Subject vs. representation (landmark vs. keychain)",
        "image1_url": "https://commons.wikimedia.org/wiki/File:Tour_Eiffel_Wikimedia_Commons.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Eiffel_Tower_Keychain.jpg",
        "answer": "No",
        "feedback": "A landmark and a keychain are not the same subject."
    },
    {
        # Scenario 5: Person vs. associated item (player vs. jersey)
        "label": "Person vs. associated item (player vs. jersey)",
        "image1_url": "https://commons.wikimedia.org/wiki/File:Lionel_Messi_in_2018.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Adidas_Messi_shirt_rear.JPG",
        "answer": "No",
        "feedback": "A person and an associated item are not the same subject."
    },
    {
        # Scenario 6: Person vs. their signature
        "label": "Person vs. their signature",
        "image1_url": "https://commons.wikimedia.org/wiki/File:President_Barack_Obama.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Health_insurance_reform_bill_signature_20100323_(1).jpg",
        "answer": "No",
        "feedback": "A person and a signature are not the same subject."
    },
    {
        # Scenario 7: Person vs. their tombstone
        "label": "Person vs. their tombstone",
        "image1_url": "https://commons.wikimedia.org/wiki/File:Oscar_Wilde_by_Napoleon_Sarony._Three-quarter-length_photograph,_seated.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Oscar_Wilde_%C3%A9_mort_dans_cette_maison.jpg",
        "answer": "No",
        "feedback": "A person and their tombstone are not the same subject."
    },
    {
        # Scenario 8: Same person at different ages (Einstein young vs old)
        "label": "Same person at different ages",
        "image1_url": "https://commons.wikimedia.org/wiki/File:JimmyCarterPortrait2.jpg",
        "image2_url": "https://commons.wikimedia.org/wiki/File:Jimmy_Carter_and_Rosalynn_Carter_on_Plains_Peanut_Festival_(cropped).jpg",
        "answer": "Yes",
        "feedback": "The same person at different ages is still the same subject."
    }
]


def check_answer(scenario_idx, user_answer):
    scenario = scenarios[scenario_idx]
    correct = scenario["answer"]
    feedback = scenario["feedback"]
    if user_answer == correct:
        return f"✅ Correct! {feedback}"
    else:
        return f"❌ Incorrect. The correct answer is **{correct}**. {feedback}"


# --- Session state init ---
if "scenario_idx" not in st.session_state:
    st.session_state.scenario_idx = 0
if "feedback" not in st.session_state:
    st.session_state.feedback = None
if "answered" not in st.session_state:
    st.session_state.answered = False

# Preload all images once at startup
if "images" not in st.session_state:
    with st.spinner("Loading all images, please wait…"):
        st.session_state.images = [
            (
                load_image(s["image1_url"]),
                load_image(s["image2_url"]),
            )
            for s in scenarios
        ]

st.title("Image Subject Comparison Tutorial")
st.markdown(
    "This is a small tutorial for Depictor. It is meant to provide examples for how to make "
    "the decisions for the questions asked."
)
st.markdown("---")
st.markdown(
    "This tutorial demonstrates rules for determining if two images depict the same subject. "
    "Answer each scenario to advance to the next one."
)

idx = st.session_state.scenario_idx
total = len(scenarios)

if idx >= total:
    st.success("🎉 You've completed all scenarios! Well done.")
    if st.button("🔄 Restart"):
        st.session_state.scenario_idx = 0
        st.session_state.feedback = None
        st.session_state.answered = False
        st.rerun()
else:
    scenario = scenarios[idx]

    st.markdown(f"### Scenario {idx + 1} of {total}: *{scenario['label']}*")
    st.progress(idx / total)

    img1, img2 = st.session_state.images[idx]
    col1, col2 = st.columns(2)
    with col1:
        if img1:
            st.image(img1, caption="Image 1", use_container_width=True)
        else:
            st.error("Image 1 could not be loaded.")
        st.markdown(f"[🔗 View source]({scenario['image1_url']})")
    with col2:
        if img2:
            st.image(img2, caption="Image 2", use_container_width=True)
        else:
            st.error("Image 2 could not be loaded.")
        st.markdown(f"[🔗 View source]({scenario['image2_url']})")

    st.markdown("### Are these the same subject?")

    if not st.session_state.answered:
        btn_col1, btn_col2 = st.columns(2)
        with btn_col1:
            if st.button("✅ Yes", use_container_width=True):
                st.session_state.feedback = check_answer(idx, "Yes")
                st.session_state.answered = True
                st.rerun()
        with btn_col2:
            if st.button("❌ No", use_container_width=True):
                st.session_state.feedback = check_answer(idx, "No")
                st.session_state.answered = True
                st.rerun()
    else:
        st.info(st.session_state.feedback)
        if st.button("➡️ Next Scenario", use_container_width=True):
            st.session_state.scenario_idx += 1
            st.session_state.feedback = None
            st.session_state.answered = False
            st.rerun()