import streamlit as st
from transformers import ViTForImageClassification, ViTFeatureExtractor
from PIL import Image
import torch
import matplotlib.pyplot as plt

# Define the repository ID
repo_id = "Hammad712/5-Flower-Types-Classification-VIT-Model"

# Load the model and feature extractor
model = ViTForImageClassification.from_pretrained(repo_id)
feature_extractor = ViTFeatureExtractor.from_pretrained(repo_id)

# Define the class names dictionary
class_names = {0: 'Lilly', 1: 'Lotus', 2: 'Orchid', 3: 'Sunflower', 4: 'Tulip'}

# Define the inference function
def predict(image):
    inputs = feature_extractor(images=image, return_tensors="pt")
    with torch.no_grad():
        outputs = model(**inputs)
    logits = outputs.logits
    probabilities = torch.nn.functional.softmax(logits, dim=-1).squeeze().tolist()
    predicted_class_idx = logits.argmax(-1).item()
    predicted_class_name = class_names[predicted_class_idx]
    return probabilities, predicted_class_name

# Streamlit app
st.title("Flower Type Classification")
st.write("Upload an image of a flower to classify its type.")

# Upload image
uploaded_file = st.sidebar.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])

if uploaded_file is not None:
    # Display the uploaded image
    image = Image.open(uploaded_file).convert("RGB")
    st.image(image, caption='Uploaded Image.', use_column_width=True)
    
    # Predict the class of the image
    probabilities, predicted_class = predict(image)
    
    # Display the probabilities in a bar chart
    fig, ax = plt.subplots()
    ax.bar(class_names.values(), probabilities)
    ax.set_ylabel('Probability')
    ax.set_xlabel('Class')
    ax.set_title('Class Probabilities')
    st.pyplot(fig)
    
    # Display the predicted class
    st.write(f"Predicted class: **{predicted_class}**")