hornet-bicho/streamlitApp

1import streamlit as st2import numpy as np3import os4import io5from PIL import Image6import tensorflow.lite as tflite7 8# Config must be the first streamlit command called9st.set_page_config(10    page_title="METEO // VISION", 11    layout="wide", 12    initial_sidebar_state="collapsed"13)14 15# Custom inject CSS for minimalist typography layout (no icons)16st.markdown("""17    <style>18    /* Global styles */19    html, body, [data-testid="stAppViewContainer"] {20        background-color: #fafafa;21        font-family: monospace;22    }23    h1, h2, h3 {24        font-family: monospace !important;25        font-weight: 700 !important;26        letter-spacing: -0.05em;27    }28    /* Style progress bars to match minimal look */29    .stProgress > div > div > div > div {30        background-color: #111111;31        border-radius: 0px;32    }33    /* Flatten buttons */34    div.stButton > button {35        border-radius: 0px !important;36        border: 1px solid #111111 !important;37        background-color: transparent;38        color: #111111;39    }40    div.stButton > button:hover {41        background-color: #111111 !important;42        color: white !important;43    }44    </style>45""", unsafe_allow_html=True)46 47# ==========================================48# 1. TFLITE MODEL LOADING & INFERENCE49# ==========================================50@st.cache_resource51def load_tflite_model(model_path):52    interpreter = tflite.Interpreter(model_path=model_path)53    interpreter.allocate_tensors()54    return interpreter55 56def predict_weather(image_bytes, interpreter):57    try:58        # Get model configurations59        input_details = interpreter.get_input_details()60        output_details = interpreter.get_output_details()61        62        input_shape = input_details[0]['shape'] 63        h, w = input_shape[1], input_shape[2]64        65        # Safely open image from raw bytes66        img = Image.open(io.BytesIO(image_bytes)).convert('RGB')67        img = img.resize((w, h))68        69        # Process arrays safely70        img_array = np.array(img, dtype=np.float32) / 255.071        img_tensor = np.expand_dims(img_array, axis=0)72        73        # Run execution engine74        interpreter.set_tensor(input_details[0]['index'], img_tensor)75        interpreter.invoke()76        77        # Extract predictions78        output_data = interpreter.get_tensor(output_details[0]['index'])[0]79        80        # Dynamically map output shape to avoid mismatch errors81        classes = ["Clear Sky", "Overcast / Cloudy", "Rainy / Stormy", "Foggy / Misty"]82        predictions = {}83        for i in range(len(output_data)):84            class_name = classes[i] if i < len(classes) else f"Class Variant {i}"85            predictions[class_name] = float(output_data[i])86            87        return predictions, None88 89    except Exception as e:90        return None, str(e)91 92# Find and dynamically bind the local file path inside the src folder environment93try:94    base_dir = os.path.dirname(os.path.abspath(__file__))95    model_path = os.path.join(base_dir, "weather_cnn.tflite")96    97    interpreter = load_tflite_model(model_path)98    model_loaded = True99except Exception as e:100    model_loaded = False101    model_error_msg = str(e)102 103# ==========================================104# 2. STYLED STREAMLIT FRONTEND105# ==========================================106 107st.title("METEO // VISION")108st.caption("Predicting localized atmospheric conditions via deep learning convolution models.")109st.write("---")110 111if not model_loaded:112    st.error(f"Failed to bind `weather_cnn.tflite` interpreter. Verification string: {model_error_msg}")113else:114    col1, col2 = st.columns(2, gap="large")115    116    with col1:117        st.subheader("VISUAL INTAKE")118        # Remove the radio buttons entirely and stick to a clean file upload box119        target_file = st.file_uploader(120            "Drop sky capture frame or photo", 121            type=["jpg", "png", "jpeg"], 122            label_visibility="collapsed"123        )124            125    with col2:126        st.subheader("METRIC DIAGNOSTICS")127        128        if target_file is not None:129            # 1. Read bytes completely upfront to stop file-pointer leakage crashes130            try:131                file_bytes = target_file.read()132                read_success = True133            except Exception as read_err:134                st.error(f"File buffer read crash: {str(read_err)}")135                read_success = False136                137            if read_success:138                with st.spinner("Processing framework matrices..."):139                    predictions, error_message = predict_weather(file_bytes, interpreter)140                    141                    if error_message:142                        st.error(f"Inference Failure: {error_message}")143                    elif predictions:144                        # Fetch dominant feature state145                        top_condition = max(predictions, key=predictions.get)146                        confidence = predictions[top_condition] * 100147                        148                        # Display structural text summary149                        st.markdown(f"## {top_condition.upper()}")150                        st.markdown(f"Confirmed with **{confidence:.1f}%** model confidence.")151                        st.write("---")152                        153                        # Output sleek probability distributions154                        for condition, score in predictions.items():155                            st.text(f"{condition:<25} {score*100:>5.1f}%")156                            st.progress(max(0.0, min(1.0, score))) # Keep between 0 and 1 bounds157        else:158            st.info("AWAITING DATA. Upload or take an environment photo to pass vectors to the CNN network layer.")