app.py

import os
import tempfile
import cv2
import streamlit as st
import PIL
import requests
from ultralytics import YOLO
import time
import numpy as np
import imageio_ffmpeg as ffmpeg
import base64

# Page config first
st.set_page_config(
    page_title="Fire Watch: Fire and Smoke Detection with an AI Vision Model",
    page_icon="🔥",
    layout="wide",
    initial_sidebar_state="expanded"
)

# Model path
model_path = 'https://huggingface.co/spaces/tstone87/ccr-colorado/resolve/main/best.pt'

# Session state initialization
for key in ["processed_frames", "slider_value", "processed_video", "start_time"]:
    if key not in st.session_state:
        st.session_state[key] = [] if key == "processed_frames" else 0 if key == "slider_value" else None

# Sidebar
with st.sidebar:
    st.header("Upload & Settings")
    source_file = st.file_uploader("Upload image/video", type=["jpg", "jpeg", "png", "bmp", "webp", "mp4"])
    confidence = float(st.slider("Confidence Threshold", 10, 100, 20)) / 100
    fps_options = {
        "Original FPS": None,
        "3 FPS": 3,
        "1 FPS": 1,
        "1 frame/4s": 0.25,
        "1 frame/10s": 0.1,
        "1 frame/15s": 0.0667,
        "1 frame/30s": 0.0333
    }
    video_option = st.selectbox("Output Frame Rate", list(fps_options.keys()))
    process_button = st.button("Detect fire")
    progress_bar = st.progress(0)
    progress_text = st.empty()
    download_slot = st.empty()

# Main page
st.title("Fire Watch: AI-Powered Fire and Smoke Detection")

# Display result images directly
col1, col2 = st.columns(2)
with col1:
    fire_4a_url = "https://huggingface.co/spaces/tstone87/ccr-colorado/resolve/main/Fire_4a.jpg"
    st.image(fire_4a_url, use_column_width=True)

with col2:
    fire_3a_url = "https://huggingface.co/spaces/tstone87/ccr-colorado/resolve/main/Fire_3a.jpg"
    st.image(fire_3a_url, use_column_width=True)

st.markdown("""
Early fire and smoke detection using YOLOv8 AI vision model. See detected results below, and upload more content for additional analysis!
""")

if not source_file:
    st.info("Please upload a file to begin.")

st.header("Your Results")
result_cols = st.columns(2)
viewer_slot = st.empty()


# Load model
try:
    model = YOLO(model_path)
except Exception as ex:
    st.error(f"Model loading failed: {str(ex)}")
    model = None

# Processing
if process_button and source_file and model:
    st.session_state.processed_frames = []
    if source_file.type.split('/')[0] == 'image':
        image = PIL.Image.open(source_file)
        res = model.predict(image, conf=confidence)
        result = res[0].plot()[:, :, ::-1]
        with result_cols[0]:
            st.image(image, caption="Original", use_column_width=True)
        with result_cols[1]:
            st.image(result, caption="Detected", use_column_width=True)
    else:
        # Video processing
        with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp:
            tmp.write(source_file.read())
            vidcap = cv2.VideoCapture(tmp.name)
        
        orig_fps = vidcap.get(cv2.CAP_PROP_FPS)
        total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
        width = int(vidcap.get(cv2.CAP_PROP_FRAME_WIDTH))
        height = int(vidcap.get(cv2.CAP_PROP_FRAME_HEIGHT))
        
        output_fps = fps_options[video_option] if fps_options[video_option] else orig_fps
        sample_interval = max(1, int(orig_fps / output_fps)) if output_fps else 1
        
        # Set fixed output FPS to 2 (500ms per frame = 2 FPS)
        fixed_output_fps = 1
        
        st.session_state.start_time = time.time()
        frame_count = 0
        processed_count = 0
        
        success, frame = vidcap.read()
        while success:
            if frame_count % sample_interval == 0:
                res = model.predict(frame, conf=confidence)
                processed_frame = res[0].plot()[:, :, ::-1]
                if not processed_frame.flags['C_CONTIGUOUS']:
                    processed_frame = np.ascontiguousarray(processed_frame)
                st.session_state.processed_frames.append(processed_frame)
                
                processed_count += 1
                elapsed = time.time() - st.session_state.start_time
                progress = frame_count / total_frames
                
                if elapsed > 0 and progress > 0:
                    total_estimated_time = elapsed / progress
                    eta = total_estimated_time - elapsed
                    elapsed_str = f"{int(elapsed // 60)}m {int(elapsed % 60)}s"
                    eta_str = f"{int(eta // 60)}m {int(eta % 60)}s" if eta > 0 else "Almost done"
                else:
                    elapsed_str = "0s"
                    eta_str = "Calculating..."
                
                progress_bar.progress(min(progress, 1.0))
                progress_text.text(f"Progress: {progress:.1%}\nElapsed: {elapsed_str}\nETA: {eta_str}")
            
            frame_count += 1
            success, frame = vidcap.read()
        
        vidcap.release()
        os.unlink(tmp.name)
        
        if st.session_state.processed_frames:
            out_path = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False).name
            writer = ffmpeg.write_frames(
                out_path,
                (width, height),
                fps=fixed_output_fps,  # Fixed at 1 FPS (1000ms per frame)
                codec='libx264',
                pix_fmt_in='bgr24',
                pix_fmt_out='yuv420p'
            )
            writer.send(None)  # Initialize writer
            
            for frame in st.session_state.processed_frames:
                writer.send(frame)
            writer.close()
            
            with open(out_path, 'rb') as f:
                st.session_state.processed_video = f.read()
            os.unlink(out_path)
            
            elapsed_final = time.time() - st.session_state.start_time
            elapsed_final_str = f"{int(elapsed_final // 60)}m {int(elapsed_final % 60)}s"
            progress_bar.progress(1.0)
            progress_text.text(f"Progress: 100%\nElapsed: {elapsed_final_str}\nETA: 0m 0s")
            with result_cols[0]:
                st.video(source_file)
            with result_cols[1]:
                st.video(st.session_state.processed_video)
            download_slot.download_button(
                label="Download Processed Video",
                data=st.session_state.processed_video,
                file_name="results_fire_analysis.mp4",
                mime="video/mp4"
            )