import streamlit as st
from PIL import Image
from transformers import pipeline
import pandas as pd

# Load calorie data
df = pd.read_csv("kaloriedata.csv")
food_list = df["navn"].tolist()

# Load zero-shot pipeline for image classification
@st.cache_resource
def get_classifier():
    return pipeline("zero-shot-image-classification", model="openai/clip-vit-base-patch32")

classifier = get_classifier()

st.title("🍽️ WebKalorier – Kalorieestimering via CLIP")

uploaded = st.file_uploader("Upload billede af mad", type=["jpg", "jpeg", "png"])
if uploaded:
    image = Image.open(uploaded).convert("RGB")
    st.image(image, caption="Uploadet billede", use_column_width=True)
    with st.spinner("Analyserer..."):
        outputs = classifier(image, candidate_labels=food_list)
    # outputs: list of dicts with labels and scores
    best = outputs[0]
    label = best['labels'][0]
    score = best['scores'][0]
    st.markdown(f"**Modelgæt:** {label} ({score:.1%} sikkerhed)")
    # fallback
    if score < 0.7:
        label = st.selectbox("Modellen er usikker – vælg manuelt:", food_list, index=food_list.index(label))
    # grams input
    gram = st.number_input(f"Angiv mængde af {label} i gram:", 1, 2000, 100)
    # lookup calories
    kcal_per_100g = df.loc[df["navn"] == label, "kcal_pr_100g"].iloc[0]
    kcal = gram * kcal_per_100g / 100
    st.subheader("🔍 Analyse af måltid")
    st.write(f"- **{gram} g {label}** → **{kcal:.1f} kcal**")
    # feedback
    feedback = st.text_input("Feedback eller korrektion (valgfrit)")
    if st.button("Send feedback"):
        with open("feedback_log.csv", "a") as f:
            f.write(f"{label},{score:.2f},{feedback}\n")
        st.success("Tak for din feedback!")