app.py

import os
import gradio as gr
import numpy as np
import math
import cv2
import tempfile
from scipy.io import loadmat, savemat
from scipy import signal, sparse, linalg
from sklearn.decomposition import FastICA
import matplotlib.pyplot as plt
import time
import mat73  # For reading SCAMPS v7.3 mat files
import pandas as pd # For LinePlot
from typing import Tuple, Dict, List # For type hinting in the new section


# --- Constants ---
MIN_BPM, MAX_BPM = 45, 180

# Fitzpatrick skin type mean RGB values (approximated from research papers)
FITZPATRICK_RGB = {
    "Type I": [239, 207, 186], "Type II": [232, 194, 163],
    "Type III": [216, 172, 134], "Type IV": [193, 142, 107],
    "Type V": [151, 103, 70], "Type VI": [82, 57, 43]
}
FITZPATRICK_TYPES = list(FITZPATRICK_RGB.keys())
SKIN_COLOR_MAP = {1: "Type I", 2: "Type II", 3: "Type III", 4: "Type IV", 5: "Type V", 6: "Type VI"}

# =================================================================================
# SECTION 1: CORE RPPG LOGIC (Used by multiple tabs)
# =================================================================================

# --- 1a. Helper Functions for Signal Processing ---
def bandpass_filter(data, fs, min_hz, max_hz):
    """Applies a Butterworth bandpass filter to the signal."""
    nyquist = 0.5 * fs; b, a = signal.butter(4, [min_hz/nyquist, max_hz/nyquist], btype='band'); return signal.filtfilt(b, a, data)

def calculate_bpm(bvp_signal, fs):
    """Calculates BPM from a BVP signal using FFT."""
    min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0; fft_data = np.abs(np.fft.rfft(bvp_signal)); freqs = np.fft.rfftfreq(len(bvp_signal), 1 / fs)
    valid_indices = np.where((freqs >= min_hz) & (freqs <= max_hz))
    if len(valid_indices[0]) == 0: return 0
    peak_freq_index = valid_indices[0][np.argmax(fft_data[valid_indices])]; peak_freq = freqs[peak_freq_index]; return peak_freq * 60

def detrend(input_signal, lambda_value):
    """Applies the smoothness priors-based detrending from the rPPG-Toolbox."""
    signal_length = input_signal.shape[0]; H = np.identity(signal_length); ones = np.ones(signal_length); minus_twos = -2 * np.ones(signal_length)
    diags_data = np.array([ones, minus_twos, ones]); diags_index = np.array([0, 1, 2])
    D = sparse.spdiags(diags_data, diags_index, (signal_length - 2), signal_length).toarray()
    if input_signal.ndim == 1: input_signal = input_signal[:, np.newaxis]
    filtered_signal = np.dot((H - linalg.inv(H + (lambda_value ** 2) * np.dot(D.T, D))), input_signal); return filtered_signal.flatten()

# --- 1b. Implementations of All Unsupervised rPPG Models ---
def rppg_green(raw_signal, fs):
    """Selects the Green channel as the BVP signal."""
    min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0; green_channel = raw_signal[:, 1]; detrended_green = signal.detrend(green_channel)
    bvp = bandpass_filter(detrended_green, fs, min_hz, max_hz); return bvp

def rppg_ica(raw_signal, fs):
    """Uses Independent Component Analysis (ICA) to separate the BVP signal."""
    normalized_signal = np.zeros_like(raw_signal)
    for i in range(raw_signal.shape[1]):
        channel_detrended = detrend(raw_signal[:, i], 100)
        normalized_signal[:, i] = (channel_detrended - np.mean(channel_detrended)) / (np.std(channel_detrended) + 1e-6)
    ica = FastICA(n_components=3, random_state=0, max_iter=1000); ica_sources = ica.fit_transform(normalized_signal)
    min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0; max_power = -np.inf; best_component = None
    for component in ica_sources.T:
        fft_data = np.abs(np.fft.rfft(component)); freqs = np.fft.rfftfreq(len(component), 1 / fs)
        valid_indices = np.where((freqs >= min_hz) & (freqs <= max_hz))
        if len(valid_indices[0]) == 0: continue
        power = np.sum(fft_data[valid_indices]**2)
        if power > max_power: max_power = power; best_component = component
    if best_component is None: raise ValueError("ICA could not find a suitable component.")
    bvp = bandpass_filter(best_component, fs, min_hz, max_hz); return bvp

def rppg_chrom(raw_signal, fs):
    """Applies the Chrominance-based method (CHROM) using a sliding window."""
    RGB = raw_signal; FN = RGB.shape[0]; WinSec = 1.6; WinL = math.ceil(WinSec * fs)
    if WinL % 2: WinL += 1
    NWin = math.floor((FN - WinL / 2) / (WinL / 2)); S = np.zeros(FN); LPF, HPF = 0.7, 2.5; NyquistF = 0.5 * fs
    B, A = signal.butter(3, [LPF / NyquistF, HPF / NyquistF], 'bandpass')
    for i in range(NWin):
        WinS = int(i * WinL / 2); WinE = int(WinS + WinL)
        if WinE > FN:
            WinE = FN
            WinL = WinE - WinS
            if WinL < 2:
                break
        RGB_win = RGB[WinS:WinE, :]; RGB_mean = np.mean(RGB_win, axis=0); RGB_norm = RGB_win / (RGB_mean + 1e-6)
        Xs = 3 * RGB_norm[:, 0] - 2 * RGB_norm[:, 1]; Ys = 1.5 * RGB_norm[:, 0] + RGB_norm[:, 1] - 1.5 * RGB_norm[:, 2]
        Xf = signal.filtfilt(B, A, Xs); Yf = signal.filtfilt(B, A, Ys); alpha = (np.std(Xf) / (np.std(Yf) + 1e-6))
        S_win = Xf - alpha * Yf; S_win = S_win * signal.windows.hann(len(S_win)); S[WinS:WinE] = S[WinS:WinE] + S_win
    return S

def rppg_pos(raw_signal, fs):
    """Applies the Plane-Orthogonal-to-Skin (POS) method using a sliding window."""
    RGB = raw_signal; N = RGB.shape[0]; H = np.zeros(N); WinSec = 1.6; l = math.ceil(WinSec * fs)
    for n in range(N):
        m = n - l + 1
        if m >= 0:
            Cn = RGB[m:n + 1, :]; mean_color = np.mean(Cn, axis=0); Cn = Cn / (mean_color + 1e-6)
            projection_matrix = np.array([[0, 1, -1], [-2, 1, 1]]); S = np.dot(Cn, projection_matrix.T)
            std_S0, std_S1 = np.std(S[:, 0]), np.std(S[:, 1]); h = S[:, 0] + (std_S0 / (std_S1 + 1e-6)) * S[:, 1]
            H[m:n + 1] = H[m:n + 1] + (h - np.mean(h))
    BVP = detrend(H, 100); min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0
    BVP = bandpass_filter(BVP, fs, min_hz, max_hz); return BVP

def rppg_lgi(raw_signal, fs):
    """Applies the Local Group Invariance (LGI) method using SVD."""
    processed_data = raw_signal.T[np.newaxis, :, :]; U, _, _ = np.linalg.svd(processed_data)
    S = U[:, :, 0]; S = np.expand_dims(S, 2); SST = np.matmul(S, np.swapaxes(S, 1, 2))
    p = np.tile(np.identity(3), (S.shape[0], 1, 1)); P = p - SST; Y = np.matmul(P, processed_data)
    bvp = Y[:, 1, :].flatten(); min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0
    bvp = bandpass_filter(bvp, fs, min_hz, max_hz); return bvp

def rppg_omit(raw_signal, fs):
    """Applies the Orthogonal Matrix Imaging Technique (OMIT) using QR decomposition."""
    processed_data = raw_signal.T; Q, R = np.linalg.qr(processed_data)
    S = Q[:, 0].reshape(1, -1); P = np.identity(3) - np.matmul(S.T, S); Y = np.dot(P, processed_data)
    bvp = Y[1, :].flatten(); min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0
    bvp = bandpass_filter(bvp, fs, min_hz, max_hz); return bvp

def rppg_pbv(raw_signal, fs):
    """Applies the Blood Volume Pulse Signature (PBV) method."""
    processed_data = raw_signal.T; sig_mean = np.mean(processed_data, axis=1, keepdims=True)
    normalized_signal = processed_data / (sig_mean + 1e-6); pbv_n = np.std(normalized_signal, axis=1)
    pbv_d = np.sqrt(np.sum(np.var(normalized_signal, axis=1))); pbv = pbv_n / (pbv_d + 1e-6)
    C = normalized_signal; Ct = C.T; Q = np.matmul(C, Ct)
    W, _, _, _ = np.linalg.lstsq(Q, pbv, rcond=None); bvp = np.matmul(Ct, W)
    min_hz, max_hz = MIN_BPM / 60.0, MAX_BPM / 60.0
    bvp = bandpass_filter(bvp, fs, min_hz, max_hz); return bvp

def rppg_additive_pos(raw_signal, fs):
    """Applies the Additive POS model based on the research paper's philosophy."""
    bvp_pos = rppg_pos(raw_signal, fs)
    normalized_signal = np.zeros_like(raw_signal)
    for i in range(raw_signal.shape[1]):
        channel_detrended = detrend(raw_signal[:, i], 100)
        normalized_signal[:, i] = (channel_detrended - np.mean(channel_detrended)) / (np.std(channel_detrended) + 1e-6)
    ica = FastICA(n_components=3, random_state=0, max_iter=1000)
    ica_sources = ica.fit_transform(normalized_signal)
    max_power = -np.inf; bvp_component_index = -1
    for i, component in enumerate(ica_sources.T):
        fft_data = np.abs(np.fft.rfft(component)); freqs = np.fft.rfftfreq(len(component), 1 / fs)
        valid_indices = np.where((freqs >= 0.75) & (freqs <= 2.5))
        if len(valid_indices[0]) == 0: continue
        power = np.sum(fft_data[valid_indices]**2)
        if power > max_power: max_power = power; bvp_component_index = i
    residual_indices = [i for i in range(3) if i != bvp_component_index]
    if len(residual_indices) < 2: return bvp_pos
    residual_signal = ica_sources[:, residual_indices[0]] + ica_sources[:, residual_indices[1]]
    peak_freq = calculate_bpm(bvp_pos, fs) / 60.0
    if peak_freq == 0: return bvp_pos
    corrected_residual = bandpass_filter(residual_signal, fs, peak_freq - 0.2, peak_freq + 0.2)
    alpha = 0.5
    norm_bvp_pos = (bvp_pos - np.mean(bvp_pos)) / (np.std(bvp_pos) + 1e-6)
    norm_corrected_residual = (corrected_residual - np.mean(corrected_residual)) / (np.std(corrected_residual) + 1e-6)
    final_bvp = norm_bvp_pos + (alpha * norm_corrected_residual)
    return final_bvp

# =================================================================================
# SECTION 2: LOGIC FOR FILE ANALYZER TAB
# =================================================================================
MODEL_DISPATCHER = {"Additive POS": rppg_additive_pos, "GREEN": rppg_green, "ICA": rppg_ica, "CHROM": rppg_chrom, "POS": rppg_pos, "LGI": rppg_lgi, "OMIT": rppg_omit, "PBV": rppg_pbv}

def load_data_from_mat(mat_file):
    """Loads data and creates a video preview from an MMPD .mat file."""
    FS = 30.0; mat_data = loadmat(mat_file.name); video_frames = mat_data['video']
    raw_signal = np.mean(video_frames, axis=(1, 2)); gt_signal = mat_data['GT_ppg'].flatten()
    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp_file: output_path = tmp_file.name
    _, height, width, _ = video_frames.shape; fourcc = cv2.VideoWriter_fourcc(*'mp4v'); out = cv2.VideoWriter(output_path, fourcc, FS, (width, height))
    for frame_float in video_frames:
        frame_uint8 = (frame_float * 255).astype(np.uint8); frame_bgr = cv2.cvtColor(frame_uint8, cv2.COLOR_RGB2BGR); out.write(frame_bgr)
    out.release(); return raw_signal, gt_signal, FS, output_path

def load_data_from_ubfc(video_file, gt_file):
    """Loads data from UBFC-format files (.avi and .txt)."""
    cap = cv2.VideoCapture(video_file.name); fs = cap.get(cv2.CAP_PROP_FPS); frames = []
    while True:
        ret, frame = cap.read()
        if not ret: break
        frames.append(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
    cap.release(); video_frames = np.array(frames); raw_signal = np.mean(video_frames, axis=(1, 2))
    with open(gt_file.name, "r") as f: gt_str = f.read().strip()
    gt_lines = gt_str.split('\n'); gt_signal = np.array([float(x) for x in gt_lines[0].split()])
    return raw_signal, gt_signal, fs, video_file.name

def load_data_from_scamps_mat(mat_file):
    """Loads data and creates a video preview from a SCAMPS .mat file."""
    FS = 30.0; mat_data = mat73.loadmat(mat_file.name)
    if 'Xsub' in mat_data: video_frames = mat_data['Xsub']
    else: raise gr.Error("SCAMPS .mat file must contain an 'Xsub' key for video.")
    if 'd_ppg' in mat_data: gt_signal = mat_data['d_ppg'].flatten()
    else: raise gr.Error("SCAMPS .mat file must contain a 'd_ppg' key for ground truth.")
    if 'fs' in mat_data: FS = mat_data['fs']
    raw_signal = np.mean(video_frames, axis=(1, 2))
    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp_file: output_path = tmp_file.name
    _, height, width, _ = video_frames.shape; fourcc = cv2.VideoWriter_fourcc(*'mp4v'); out = cv2.VideoWriter(output_path, fourcc, FS, (width, height))
    for frame_float in video_frames:
        frame_uint8 = (frame_float * 255).astype(np.uint8); frame_bgr = cv2.cvtColor(frame_uint8, cv2.COLOR_RGB2BGR); out.write(frame_bgr)
    out.release(); return raw_signal, gt_signal, FS, output_path

def process_file_and_evaluate(dataset_type, mat_file_mmpd, ubfc_vid_file, ubfc_gt_file, mat_file_scamps, selected_model):
    """Main processing pipeline for all file-based analysis."""
    if dataset_type == "MMPD":
        if mat_file_mmpd is None: raise gr.Error("Please upload an MMPD .mat file.")
        raw_signal, gt_signal, fs, video_path = load_data_from_mat(mat_file_mmpd)
    elif dataset_type == "UBFC":
        if ubfc_vid_file is None or ubfc_gt_file is None: raise gr.Error("Please upload both a video and a ground truth file for UBFC.")
        raw_signal, gt_signal, fs, video_path = load_data_from_ubfc(ubfc_vid_file, ubfc_gt_file)
    elif dataset_type == "SCAMPS":
        if mat_file_scamps is None: raise gr.Error("Please upload a SCAMPS .mat file.")
        raw_signal, gt_signal, fs, video_path = load_data_from_scamps_mat(mat_file_scamps)
    else: raise gr.Error("Invalid dataset type selected.")
    
    if len(gt_signal) != len(raw_signal): gt_signal = signal.resample(gt_signal, len(raw_signal))
    
    model_function = MODEL_DISPATCHER[selected_model]; predicted_bvp = model_function(raw_signal, fs)
    predicted_bpm = calculate_bpm(predicted_bvp, fs); gt_bpm = calculate_bpm(gt_signal, fs)
    
    fig = plt.figure(figsize=(12, 6))
    time_values = np.arange(len(gt_signal)) / fs
    norm_predicted_bvp = (predicted_bvp - np.mean(predicted_bvp)) / (np.std(predicted_bvp) + 1e-6)
    norm_gt_signal = (gt_signal - np.mean(gt_signal)) / (np.std(gt_signal) + 1e-6)
    plt.plot(time_values, norm_gt_signal, label=f'Ground Truth BVP (BPM: {gt_bpm:.2f})', alpha=0.8)
    plt.plot(time_values, norm_predicted_bvp, label=f'Predicted BVP ({selected_model}) (BPM: {predicted_bpm:.2f})', linestyle='--')
    plt.title("Ground Truth vs. Predicted BVP Signal"); plt.xlabel("Time (seconds)"), plt.ylabel("Normalized Amplitude"), plt.grid(True), plt.legend(), plt.tight_layout()
    
    mae = abs(predicted_bpm - gt_bpm); mape = (mae / gt_bpm) * 100 if gt_bpm != 0 else float('inf')
    eval_results = (f"Ground Truth BPM: {gt_bpm:.2f}\nPredicted BPM: {predicted_bpm:.2f}\n"
                    f"Mean Absolute Error (MAE): {mae:.2f}\nMean Absolute Percentage Error (MAPE): {mape:.2f} %")
    
    plt.close(fig) # Fix for memory leak
    return f"{predicted_bpm:.2f} BPM", fig, eval_results, video_path

# =================================================================================
# SECTION 3: LOGIC FOR LIVE WEBCAM PREDICTION TAB
# =================================================================================
BUFFER_SIZE = 300; FS_WEBCAM = 30; face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml'); DETECTION_INTERVAL = 5

def analyze_bvp_signal(bvp_signal, fs):
    """Takes a BVP signal and returns HR, HRV, estimated BP, and plot data."""
    hr = calculate_bpm(bvp_signal, fs)
    try:
        peaks, _ = signal.find_peaks(bvp_signal, height=np.mean(bvp_signal), distance=fs * 0.5)
        if len(peaks) > 1:
            rr_intervals = np.diff(peaks) / fs * 1000
            successive_diffs = np.diff(rr_intervals)
            hrv_rmssd = np.sqrt(np.mean(successive_diffs ** 2))
        else:
            hrv_rmssd = 0
    except Exception:
        hrv_rmssd = 0
    systolic_bp = (0.5 * hr) + 90; diastolic_bp = (0.3 * hr) + 60
    time_values = np.arange(len(bvp_signal)) / fs
    plot_df = pd.DataFrame({"time": time_values, "BVP": bvp_signal})
    return hr, hrv_rmssd, systolic_bp, diastolic_bp, plot_df

def process_webcam_frame(frame, signal_buffer, last_update_time, selected_model, frame_counter, last_face_box):
    """Processes each frame from the webcam for live prediction."""
    if frame is None:
        return gr.update(), gr.update(), gr.update(), gr.update(), gr.update(), None, signal_buffer, last_update_time, frame_counter, last_face_box
    
    frame_with_feedback = frame.copy(); frame_counter += 1; face_box = None
    
    if frame_counter % DETECTION_INTERVAL == 0:
        gray = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY); faces = face_cascade.detectMultiScale(gray, 1.1, 5)
        if len(faces) > 0:
            face_box = faces[0]
            last_face_box = face_box
    else:
        face_box = last_face_box
    status_text = "Detecting face..."
    if face_box is not None:
        x, y, w, h = face_box; cv2.rectangle(frame_with_feedback, (x, y), (x + w, y + h), (0, 255, 0), 2)
        forehead_y, forehead_h = y + int(0.08 * h), int(0.18 * h); forehead_x, forehead_w = x + int(0.25 * w), int(0.5 * w)
        roi = frame[forehead_y:forehead_y + forehead_h, forehead_x:forehead_x + forehead_w]
        if roi.size > 0:
            avg_rgb = np.mean(roi, axis=(0, 1)); signal_buffer.append(avg_rgb)
            if len(signal_buffer) > BUFFER_SIZE:
                signal_buffer.pop(0)
            progress = int((len(signal_buffer) / BUFFER_SIZE) * 100); status_text = f"Collecting data... ({progress}%)"
    else:
        status_text = "Face not detected..."
    
    current_time = time.time()
    if len(signal_buffer) == BUFFER_SIZE and (current_time - last_update_time) > 1:
        model_function = MODEL_DISPATCHER[selected_model]; bvp_signal = model_function(np.array(signal_buffer), FS_WEBCAM)
        hr, hrv, sbp, dbp, plot_data = analyze_bvp_signal(bvp_signal, FS_WEBCAM)
        hr_text = f"{hr:.2f} BPM"; hrv_text = f"{hrv:.2f} ms (RMSSD)"; bp_text = f"{sbp:.1f} / {dbp:.1f} mmHg (est.)"
        return status_text, hr_text, hrv_text, bp_text, plot_data, frame_with_feedback, signal_buffer, current_time, frame_counter, last_face_box
    
    return status_text, gr.update(), gr.update(), gr.update(), gr.update(), frame_with_feedback, signal_buffer, last_update_time, frame_counter, last_face_box

# =================================================================================
# SECTION 4: LOGIC FOR SKIN COLOR MANIPULATION TAB
# =================================================================================
def get_skin_mask(frame_rgb):
    """Creates a binary mask to identify skin pixels."""
    frame_hsv = cv2.cvtColor(frame_rgb, cv2.COLOR_RGB2HSV); lower_hsv = np.array([0, 48, 80], dtype="uint8")
    upper_hsv = np.array([20, 255, 255], dtype="uint8"); skin_mask = cv2.inRange(frame_hsv, lower_hsv, upper_hsv); return skin_mask

def get_skin_info_and_preview(mat_file):
    """Loads an MMPD file, identifies its skin type, and creates a video preview."""
    if mat_file is None: return "N/A", gr.Dropdown(choices=FITZPATRICK_TYPES), None
    try:
        mat_data = loadmat(mat_file.name); original_skin_type_id = mat_data['skin_color'].item()
        original_skin_type_name = SKIN_COLOR_MAP.get(original_skin_type_id, "Unknown")
        dropdown_choices = [ftype for ftype in FITZPATRICK_TYPES if ftype != original_skin_type_name]
        _, _, _, video_path = load_data_from_mat(mat_file)
        return original_skin_type_name, gr.Dropdown(choices=dropdown_choices, value=dropdown_choices[0]), video_path
    except Exception as e: raise gr.Error(f"Failed to process .mat file: {e}")

def manipulate_and_compare(mat_file, target_skin_type_name, selected_model):
    """Manipulates skin color and performs a comparative analysis."""
    if mat_file is None or not target_skin_type_name or not selected_model: raise gr.Error("Please upload a file and select all options.")
    mat_data = loadmat(mat_file.name); original_video_frames_float = mat_data['video']
    original_video_frames_uint8 = (original_video_frames_float * 255).astype(np.uint8)
    gt_signal = mat_data['GT_ppg'].flatten(); fs = 30.0
    original_raw_signal = np.mean(original_video_frames_float, axis=(1, 2))
    if len(gt_signal) != len(original_raw_signal): gt_signal = signal.resample(gt_signal, len(original_raw_signal))
    gt_bpm = calculate_bpm(gt_signal, fs); model_function = MODEL_DISPATCHER[selected_model]
    original_bvp = model_function(original_raw_signal, fs); original_bpm = calculate_bpm(original_bvp, fs); original_mae = abs(original_bpm - gt_bpm)
    target_color = np.array(FITZPATRICK_RGB[target_skin_type_name]); first_frame = original_video_frames_uint8[0]
    gray = cv2.cvtColor(first_frame, cv2.COLOR_RGB2GRAY); faces = face_cascade.detectMultiScale(gray, 1.1, 5)
    if len(faces) == 0: raise gr.Error("Could not detect a face in the first frame.")
    x, y, w, h = faces[0]; forehead_roi = first_frame[y + int(0.08*h) : y + int(0.26*h), x + int(0.25*w) : x + int(0.75*w)]
    source_color = np.mean(forehead_roi, axis=(0, 1)); manipulated_frames = []; color_offset = target_color - source_color
    for frame in original_video_frames_uint8:
        skin_mask = get_skin_mask(frame); skin_mask_3ch = cv2.cvtColor(skin_mask, cv2.COLOR_GRAY2RGB)
        offset_image = (skin_mask_3ch / 255 * color_offset).astype('int16')
        manipulated_frame = np.clip(frame.astype('int16') + offset_image, 0, 255).astype('uint8'); manipulated_frames.append(manipulated_frame)
    manipulated_video_float = np.array(manipulated_frames) / 255.0
    manipulated_raw_signal = np.mean(manipulated_video_float, axis=(1, 2))
    manipulated_bvp = model_function(manipulated_raw_signal, fs); manipulated_bpm = calculate_bpm(manipulated_bvp, fs); manipulated_mae = abs(manipulated_bpm - gt_bpm)
    comparison_text = (f"--- Analysis using {selected_model} ---\n\n"
                       f"Ground Truth BPM: {gt_bpm:.2f}\n\n"
                       f"Original Video:\n  - Predicted BPM: {original_bpm:.2f}\n  - MAE: {original_mae:.2f}\n\n"
                       f"Manipulated Video ({target_skin_type_name}):\n  - Predicted BPM: {manipulated_bpm:.2f}\n  - MAE: {manipulated_mae:.2f}")
    fig = plt.figure(figsize=(12, 6))
    time_values = np.arange(len(gt_signal)) / fs
    norm_gt = (gt_signal - np.mean(gt_signal)) / (np.std(gt_signal) + 1e-6)
    norm_orig = (original_bvp - np.mean(original_bvp)) / (np.std(original_bvp) + 1e-6)
    norm_manip = (manipulated_bvp - np.mean(manipulated_bvp)) / (np.std(manipulated_bvp) + 1e-6)
    plt.plot(time_values, norm_gt, label=f'Ground Truth (BPM: {gt_bpm:.2f})', alpha=0.9)
    plt.plot(time_values, norm_orig, label=f'Original Pred. (BPM: {original_bpm:.2f})', linestyle='--')
    plt.plot(time_values, norm_manip, label=f'Manipulated Pred. (BPM: {manipulated_bpm:.2f})', linestyle=':')
    plt.title("BVP Signal Comparison"); plt.xlabel("Time (seconds)"); plt.ylabel("Normalized Amplitude"); plt.grid(True); plt.legend(); plt.tight_layout()
    new_mat_data = mat_data.copy(); new_mat_data['video'] = manipulated_video_float
    new_skin_type_id = FITZPATRICK_TYPES.index(target_skin_type_name) + 1; new_mat_data['skin_color'] = np.array([[new_skin_type_id]])
    with tempfile.NamedTemporaryFile(suffix=".mat", delete=False) as tmp_file: new_mat_path = tmp_file.name
    savemat(new_mat_path, new_mat_data, do_compression=True)
    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp_file: video_out_path = tmp_file.name
    _, height, width, _ = original_video_frames_uint8.shape
    fourcc = cv2.VideoWriter_fourcc(*'mp4v'); out = cv2.VideoWriter(video_out_path, fourcc, fs, (width, height))
    for frame in manipulated_frames: out.write(cv2.cvtColor(frame, cv2.COLOR_RGB2BGR))
    out.release()
    plt.close(fig) # Fix for memory leak
    return video_out_path, new_mat_path, comparison_text, fig

# =================================================================================
# SECTION 5: LOGIC FOR DATASET CONVERTER TAB
# =================================================================================
def convert_ubfc_to_mmpd(video_file, gt_file, resize_option):
    """Converts UBFC files to a single MMPD-compatible .mat file."""
    if video_file is None or gt_file is None:
        raise gr.Error("Please upload both UBFC video and ground truth files.")
    cap = cv2.VideoCapture(video_file.name); frames = []
    fs = cap.get(cv2.CAP_PROP_FPS)
    message = "Conversion successful!"
    while True:
        ret, frame = cap.read()
        if not ret: break
        if resize_option == "Downsample to 320p (Recommended)":
            target_width = 320
            original_width = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
            original_height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
            target_height = int(target_width * (original_height / original_width))
            processed_frame = cv2.resize(frame, (target_width, target_height))
            message = f"Conversion successful! Video resized to {target_width}x{target_height}."
        else:
            processed_frame = frame
        frames.append(cv2.cvtColor(processed_frame, cv2.COLOR_BGR2RGB) / 255.0)
    cap.release(); video_frames_float = np.array(frames, dtype=np.float32)
    with open(gt_file.name, "r") as f: gt_str = f.read().strip()
    gt_lines = gt_str.split('\n'); gt_signal = np.array([float(x) for x in gt_lines[0].split()])
    mmpd_data = {
        'video': video_frames_float, 'GT_ppg': gt_signal, 'skin_color': np.array([[3]]), 'gender': np.array([['male']]),
        'light': np.array([['LED-low']]), 'motion': np.array([['Stationary']]), 'exercise': np.array([['False']]),
        'glasser': np.array([['False']]), 'hair_cover': np.array([['False']]), 'makeup': np.array([['False']]), 'fps': np.array([[fs]])
    }
    with tempfile.NamedTemporaryFile(suffix=".mat", delete=False) as tmp_file: new_mat_path = tmp_file.name
    savemat(new_mat_path, mmpd_data, do_compression=True)
    return new_mat_path, message

def convert_scamps_to_mmpd(mat_file, resize_option):
    """Converts a SCAMPS .mat file to an MMPD-compatible .mat file."""
    if mat_file is None: raise gr.Error("Please upload a SCAMPS .mat file.")
    mat_data = mat73.loadmat(mat_file.name)
    if 'Xsub' in mat_data: video_frames = mat_data['Xsub']
    else: raise gr.Error("SCAMPS .mat file must contain an 'Xsub' key.")
    if 'd_ppg' in mat_data: gt_signal = mat_data['d_ppg'].flatten()
    else: raise gr.Error("SCAMPS .mat file must contain a 'd_ppg' key.")
    fs = mat_data.get('fs', 30.0)
    message = "Conversion successful!"
    if resize_option == "Downsample to 320p (Recommended)":
        num_frames, h, w, _ = video_frames.shape
        target_width = 320; target_height = int(target_width * (h / w))
        resized_frames = np.zeros((num_frames, target_height, target_width, 3), dtype=np.float32)
        for i, frame in enumerate(video_frames):
            resized_frames[i] = cv2.resize(frame, (target_width, target_height), interpolation=cv2.INTER_AREA)
        video_frames_float = resized_frames
        message = f"Conversion successful! Video resized to {target_width}x{target_height}."
    else:
        video_frames_float = np.array(video_frames, dtype=np.float32)
    mmpd_data = {
        'video': video_frames_float, 'GT_ppg': gt_signal, 'skin_color': np.array([[3]]), 'gender': np.array([['male']]),
        'light': np.array([['LED-low']]), 'motion': np.array([['Stationary']]), 'exercise': np.array([['False']]),
        'glasser': np.array([['False']]), 'hair_cover': np.array([['False']]), 'makeup': np.array([['False']]), 'fps': np.array([[fs]])
    }
    with tempfile.NamedTemporaryFile(suffix=".mat", delete=False) as tmp_file: new_mat_path = tmp_file.name
    savemat(new_mat_path, mmpd_data, do_compression=True)
    return new_mat_path, message

# =================================================================================
# SECTION 6: LOGIC FOR SYNTHETIC DATASET GENERATOR TAB
# =================================================================================
class CameraConditionSimulator:
    def __init__(self, seed: int = 42):
        np.random.seed(seed)
    def add_sensor_noise(self, frame: np.ndarray, noise_level: str = 'medium') -> np.ndarray:
        noise_params = {'low': {'shot_sigma': 2, 'read_sigma': 1}, 'medium': {'shot_sigma': 5, 'read_sigma': 3}, 'high': {'shot_sigma': 10, 'read_sigma': 5}}
        params = noise_params[noise_level]
        shot_noise = np.random.normal(0, params['shot_sigma'], frame.shape) * np.sqrt(frame / 255.0)
        read_noise = np.random.normal(0, params['read_sigma'], frame.shape)
        noisy_frame = frame + shot_noise + read_noise
        return np.clip(noisy_frame, 0, 255).astype(np.uint8)
    def add_motion_blur(self, frame: np.ndarray, intensity: float = 0.5) -> np.ndarray:
        kernel_size = int(5 * intensity) * 2 + 1
        if kernel_size < 3: return frame
        kernel = np.zeros((kernel_size, kernel_size)); kernel[kernel_size // 2, :] = np.ones(kernel_size); kernel = kernel / kernel_size
        return cv2.filter2D(frame, -1, kernel).astype(np.uint8)
    def simulate_compression(self, frame: np.ndarray, quality: int = 25) -> np.ndarray:
        frame_bgr = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
        encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), quality]; _, encoded = cv2.imencode('.jpg', frame_bgr, encode_param)
        return cv2.imdecode(encoded, cv2.IMREAD_COLOR)
    def adjust_lighting(self, frame: np.ndarray, illumination: int = 300) -> np.ndarray:
        scale_factor = illumination / 300; gamma = 1.0 if scale_factor >= 1.0 else 0.8
        adjusted = np.power(frame.astype(np.float32) * scale_factor / 255.0, gamma) * 255.0
        if illumination < 200: adjusted = self.add_sensor_noise(adjusted, 'high')
        elif illumination < 300: adjusted = self.add_sensor_noise(adjusted, 'medium')
        return np.clip(adjusted, 0, 255).astype(np.uint8)

def generate_synthetic_dataset(mat_file, resolution_key, environment_key, selected_model):
    if mat_file is None: raise gr.Error("Please upload an MMPD .mat file first.")
    
    resolution_configs = {'480p': {'resolution': (640, 480), 'fps': 30, 'noise_level': 'medium', 'compression_quality': 28}, '720p': {'resolution': (1280, 720), 'fps': 30, 'noise_level': 'medium', 'compression_quality': 25}, '1080p30': {'resolution': (1920, 1080), 'fps': 30, 'noise_level': 'low', 'compression_quality': 23}}
    environment_presets = {'optimal': {'illumination': 300, 'motion_blur': 0.0}, 'low_light': {'illumination': 150, 'motion_blur': 0.0}, 'motion': {'illumination': 300, 'motion_blur': 0.5}}

    res_config = resolution_configs[resolution_key]
    env_config = environment_presets[environment_key]
    
    mat_data = loadmat(mat_file.name)
    original_video_frames_float = mat_data['video']
    gt_signal = mat_data['GT_ppg'].flatten()
    fs = mat_data.get('fps', np.array([[30.0]])).item()
    
    original_raw_signal = np.mean(original_video_frames_float, axis=(1, 2))
    if len(gt_signal) != len(original_raw_signal): gt_signal = signal.resample(gt_signal, len(original_raw_signal))
    gt_bpm = calculate_bpm(gt_signal, fs)
    model_function = MODEL_DISPATCHER[selected_model]
    original_bvp = model_function(original_raw_signal, fs)
    original_bpm = calculate_bpm(original_bvp, fs)
    original_mae = abs(original_bpm - gt_bpm)

    simulator = CameraConditionSimulator()
    new_frames_uint8_bgr = []
    
    for frame_float in original_video_frames_float:
        frame_uint8_rgb = (frame_float * 255).astype(np.uint8)
        resized_rgb = cv2.resize(frame_uint8_rgb, res_config['resolution'], interpolation=cv2.INTER_AREA)
        processed_rgb = simulator.adjust_lighting(resized_rgb, env_config['illumination'])
        if env_config['motion_blur'] > 0: processed_rgb = simulator.add_motion_blur(processed_rgb, env_config['motion_blur'])
        processed_rgb = simulator.add_sensor_noise(processed_rgb, res_config['noise_level'])
        compressed_bgr = simulator.simulate_compression(processed_rgb, res_config['compression_quality'])
        new_frames_uint8_bgr.append(compressed_bgr)

    new_video_float_rgb = np.array([cv2.cvtColor(f, cv2.COLOR_BGR2RGB) for f in new_frames_uint8_bgr], dtype=np.float32) / 255.0
    synthetic_raw_signal = np.mean(new_video_float_rgb, axis=(1, 2))
    synthetic_bvp = model_function(synthetic_raw_signal, fs)
    synthetic_bpm = calculate_bpm(synthetic_bvp, fs)
    synthetic_mae = abs(synthetic_bpm - gt_bpm)

    comparison_text = (f"--- Analysis using {selected_model} ---\n\n"
                       f"Ground Truth BPM: {gt_bpm:.2f}\n\n"
                       f"Original Video:\n  - Predicted BPM: {original_bpm:.2f}\n  - MAE: {original_mae:.2f}\n\n"
                       f"Synthetic Video ({resolution_key}, {environment_key}):\n  - Predicted BPM: {synthetic_bpm:.2f}\n  - MAE: {synthetic_mae:.2f}")

    fig = plt.figure(figsize=(12, 6))
    time_values = np.arange(len(gt_signal)) / fs
    norm_gt = (gt_signal - np.mean(gt_signal)) / (np.std(gt_signal) + 1e-6)
    norm_orig = (original_bvp - np.mean(original_bvp)) / (np.std(original_bvp) + 1e-6)
    norm_synth = (synthetic_bvp - np.mean(synthetic_bvp)) / (np.std(synthetic_bvp) + 1e-6)
    plt.plot(time_values, norm_gt, label=f'Ground Truth (BPM: {gt_bpm:.2f})', alpha=0.9)
    plt.plot(time_values, norm_orig, label=f'Original Pred. (BPM: {original_bpm:.2f})', linestyle='--')
    plt.plot(time_values, norm_synth, label=f'Synthetic Pred. (BPM: {synthetic_bpm:.2f})', linestyle=':')
    plt.title("BVP Signal Comparison"); plt.xlabel("Time (seconds)"); plt.ylabel("Normalized Amplitude"); plt.grid(True); plt.legend(); plt.tight_layout()
    
    num_frames, h, w, _ = new_video_float_rgb.shape
    target_width = 320
    target_height = int(target_width * (h / w))
    resized_frames_for_mat = np.zeros((num_frames, target_height, target_width, 3), dtype=np.float32)
    for i, frame in enumerate(new_video_float_rgb):
        frame_uint8 = (frame * 255).astype(np.uint8)
        resized_frame = cv2.resize(frame_uint8, (target_width, target_height), interpolation=cv2.INTER_AREA)
        resized_frames_for_mat[i] = resized_frame.astype(np.float32) / 255.0

    new_mat_data = mat_data.copy(); new_mat_data['video'] = resized_frames_for_mat
    new_mat_data['fps'] = np.array([[res_config['fps']]])
    with tempfile.NamedTemporaryFile(suffix=".mat", delete=False) as tmp_file: new_mat_path = tmp_file.name
    savemat(new_mat_path, new_mat_data, do_compression=True)
    
    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp_file: video_out_path = tmp_file.name
    fourcc = cv2.VideoWriter_fourcc(*'mp4v'); out = cv2.VideoWriter(video_out_path, fourcc, res_config['fps'], res_config['resolution'])
    for frame in new_frames_uint8_bgr:
        out.write(frame)
    out.release()

    plt.close(fig)
    return video_out_path, new_mat_path, comparison_text, fig

# =================================================================================
# SECTION 7: GRADIO USER INTERFACE
# =================================================================================
with gr.Blocks(theme=gr.themes.Soft(), title="rPPG Analysis and Prediction Toolbox") as demo:
    gr.Markdown("# rPPG Analysis and Prediction Toolbox\n*Idea taken from the [rPPG-Toolbox](https://github.com/ubicomplab/rPPG-Toolbox)*")
    
    with gr.Tabs():
        with gr.TabItem("Live Webcam Prediction"):
            with gr.Row():
                with gr.Column():
                    gr.Markdown("## 1. Live Camera Feed"); webcam_component = gr.Image(sources=["webcam"], streaming=True, label="Webcam Feed with Face Detection")
                    model_selector_live = gr.Dropdown(choices=["Additive POS", "POS", "CHROM", "ICA", "GREEN", "LGI", "OMIT", "PBV"], value="Additive POS", label="Select rPPG Model for Live Analysis")
                with gr.Column():
                    gr.Markdown("## 2. Real-Time Results")
                    status_live_text = gr.Textbox(label="Status", interactive=False)
                    hr_live_output = gr.Textbox(label="Heart Rate (HR)")
                    hrv_live_output = gr.Textbox(label="Heart Rate Variability (HRV)")
                    bp_live_output = gr.Textbox(label="Blood Pressure (BP) Estimation")
                    gr.Markdown("*Disclaimer: Blood pressure estimation is for demonstration purposes only and is not medically accurate.*")
                    bvp_plot_live = gr.LinePlot(x="time", y="BVP", title="Live BVP Signal", show_label=False, width=400, height=300)
            signal_buffer_state = gr.State([]); last_update_time_state = gr.State(0)
            frame_counter_state = gr.State(0); last_face_box_state = gr.State(None)
            webcam_component.stream(fn=process_webcam_frame, inputs=[webcam_component, signal_buffer_state, last_update_time_state, model_selector_live, frame_counter_state, last_face_box_state], outputs=[status_live_text, hr_live_output, hrv_live_output, bp_live_output, bvp_plot_live, webcam_component, signal_buffer_state, last_update_time_state, frame_counter_state, last_face_box_state])

        with gr.TabItem("File Analyzer"):
            with gr.Row():
                with gr.Column(scale=1):
                    gr.Markdown("## 1. Inputs"); video_output_file = gr.Video(label="Video Preview", interactive=False)
                    dataset_type_selector = gr.Radio(choices=["MMPD", "UBFC", "SCAMPS"], value="MMPD", label="Select Dataset Format")
                    with gr.Group(visible=True) as mmpd_uploader: mat_input_mmpd = gr.File(label="Upload .mat File (MMPD Format)", file_types=[".mat"])
                    with gr.Group(visible=False) as ubfc_uploader:
                        ubfc_vid_input = gr.File(label="Upload video.avi File", file_types=["video"]); ubfc_gt_input = gr.File(label="Upload ground_truth.txt", file_types=[".txt"])
                    with gr.Group(visible=False) as scamps_uploader:
                        mat_input_scamps = gr.File(label="Upload .mat File (SCAMPS Format)", file_types=[".mat"])
                    model_selector_file = gr.Dropdown(choices=["Additive POS", "POS", "CHROM", "ICA", "GREEN", "LGI", "OMIT", "PBV"], value="Additive POS", label="Select rPPG Model")
                    run_button_file = gr.Button("Analyze & Evaluate", variant="primary")
                with gr.Column(scale=2):
                    gr.Markdown("## 2. Results"); bpm_output_file = gr.Textbox(label="Predicted Heart Rate (BPM)")
                    bvp_plot_file = gr.Plot(label="BVP Signal Comparison"); eval_output_file = gr.Textbox(label="Evaluation Metrics", lines=4)
            def switch_uploader(dataset_type):
                if dataset_type == "MMPD": return gr.Group(visible=True), gr.Group(visible=False), gr.Group(visible=False)
                elif dataset_type == "UBFC": return gr.Group(visible=False), gr.Group(visible=True), gr.Group(visible=False)
                else: return gr.Group(visible=False), gr.Group(visible=False), gr.Group(visible=True)
            dataset_type_selector.change(fn=switch_uploader, inputs=dataset_type_selector, outputs=[mmpd_uploader, ubfc_uploader, scamps_uploader])
            run_button_file.click(fn=process_file_and_evaluate, inputs=[dataset_type_selector, mat_input_mmpd, ubfc_vid_input, ubfc_gt_input, mat_input_scamps, model_selector_file], outputs=[bpm_output_file, bvp_plot_file, eval_output_file, video_output_file])

        with gr.TabItem("Skin Color Manipulation"):
            with gr.Row():
                with gr.Column():
                    gr.Markdown("## 1. Upload & Select"); mat_input_skin = gr.File(label="Upload .mat File (MMPD Format)", file_types=[".mat"])
                    original_skin_type_text = gr.Textbox(label="Original Skin Type", interactive=False); target_skin_type_dropdown = gr.Dropdown(label="Select Target Skin Type")
                    model_selector_skin = gr.Dropdown(choices=["Additive POS", "POS", "CHROM", "ICA", "GREEN", "LGI", "OMIT", "PBV"], value="Additive POS", label="Select rPPG Model for Comparison")
                    manipulate_button = gr.Button("Manipulate & Compare", variant="primary")
                with gr.Column():
                    gr.Markdown("## 2. Preview, Compare & Download")
                    with gr.Row():
                        video_preview_original = gr.Video(label="Original Video", interactive=False); video_preview_manipulated = gr.Video(label="Manipulated Video", interactive=False)
                    comparison_results_text = gr.Textbox(label="Comparison Results", lines=8)
                    comparison_plot = gr.Plot(label="BVP Signal Comparison Plot")
                    download_button = gr.File(label="Download Manipulated .mat File", interactive=True)
            mat_input_skin.upload(fn=get_skin_info_and_preview, inputs=mat_input_skin, outputs=[original_skin_type_text, target_skin_type_dropdown, video_preview_original])
            manipulate_button.click(fn=manipulate_and_compare, inputs=[mat_input_skin, target_skin_type_dropdown, model_selector_skin], outputs=[video_preview_manipulated, download_button, comparison_results_text, comparison_plot])

        with gr.TabItem("Resolution and Environment Converter"):
            with gr.Row():
                with gr.Column():
                    gr.Markdown("## 1. Upload & Configure"); synth_mat_input = gr.File(label="Upload .mat File (MMPD Format)", file_types=[".mat"])
                    original_resolution_text = gr.Textbox(label="Original Video Resolution", interactive=False)
                    synth_resolution_dd = gr.Dropdown(label="Target Resolution", choices=['480p', '720p', '1080p30'], value='720p')
                    synth_env_dd = gr.Dropdown(label="Target Environment", choices=['optimal', 'low_light', 'motion'], value='optimal')
                    model_selector_synth = gr.Dropdown(choices=["Additive POS", "POS", "CHROM", "ICA", "GREEN", "LGI", "OMIT", "PBV"], value="Additive POS", label="Select rPPG Model for Comparison")
                    synth_generate_btn = gr.Button("Generate & Compare", variant="primary")
                with gr.Column():
                    gr.Markdown("## 2. Preview, Compare & Download")
                    with gr.Row():
                        synth_video_original = gr.Video(label="Original Video", interactive=False); synth_video_generated = gr.Video(label="Generated Synthetic Video", interactive=False)
                    synth_comparison_text = gr.Textbox(label="Comparison Results", lines=8)
                    synth_comparison_plot = gr.Plot(label="BVP Signal Comparison Plot")
                    synth_download_btn = gr.File(label="Download Generated .mat File", interactive=True)
            def synth_get_info_and_preview(mat_file):
                if mat_file is None: return None, "N/A", gr.Dropdown(choices=['480p', '720p', '1080p30'])
                _, _, _, video_path = load_data_from_mat(mat_file)
                mat_data = loadmat(mat_file.name)
                h, w = mat_data['video'].shape[1], mat_data['video'].shape[2]
                original_res_str = f"{w}x{h}"
                all_resolutions = {'480p': 480, '720p': 720, '1080p30': 1080}
                original_res_key = None
                for key, val in all_resolutions.items():
                    if abs(h - val) < 50: original_res_key = key; break
                new_choices = [res for res in all_resolutions.keys() if res != original_res_key]
                return video_path, original_res_str, gr.Dropdown(choices=new_choices, value=new_choices[0] if new_choices else None)
            synth_mat_input.upload(fn=synth_get_info_and_preview, inputs=synth_mat_input, outputs=[synth_video_original, original_resolution_text, synth_resolution_dd])
            synth_generate_btn.click(fn=generate_synthetic_dataset, inputs=[synth_mat_input, synth_resolution_dd, synth_env_dd, model_selector_synth], outputs=[synth_video_generated, synth_download_btn, synth_comparison_text, synth_comparison_plot])
        
        with gr.TabItem("Dataset Converter"):
            with gr.Column():
                gr.Markdown("## 1. Select Conversion Type")
                conversion_type_selector = gr.Radio(choices=["UBFC to MMPD", "SCAMPS to MMPD"], value="UBFC to MMPD", label="Select Conversion")
                with gr.Group(visible=True) as ubfc_converter_group:
                    gr.Markdown("### Upload UBFC Files")
                    ubfc_conv_vid = gr.File(label="Upload video.avi File", file_types=["video"])
                    ubfc_conv_gt = gr.File(label="Upload ground_truth.txt", file_types=[".txt"])
                    ubfc_resize_option = gr.Radio(choices=["Downsample to 320p (Recommended)", "Keep Original Size (May Fail for Large Files)"], value="Downsample to 320p (Recommended)", label="Video Size Option")
                    gr.Markdown("*Warning: Keeping original size may cause an OverflowError if the video file is very large.*")
                    convert_button_ubfc = gr.Button("Convert UBFC to MMPD", variant="primary")
                with gr.Group(visible=False) as scamps_converter_group:
                    gr.Markdown("### Upload SCAMPS File")
                    scamps_conv_mat = gr.File(label="Upload .mat File (SCAMPS Format)", file_types=[".mat"])
                    scamps_resize_option = gr.Radio(choices=["Downsample to 320p (Recommended)", "Keep Original Size (May Fail for Large Files)"], value="Downsample to 320p (Recommended)", label="Video Size Option")
                    gr.Markdown("*Warning: Keeping original size may cause an OverflowError if the video file is very large.*")
                    convert_button_scamps = gr.Button("Convert SCAMPS to MMPD", variant="primary")
                gr.Markdown("## 2. Download Converted File")
                status_text_conv = gr.Textbox(label="Status", interactive=False)
                download_button_conv = gr.File(label="Download Converted .mat File", interactive=True)
            def switch_converter(conv_type):
                if conv_type == "UBFC to MMPD": return gr.Group(visible=True), gr.Group(visible=False)
                else: return gr.Group(visible=False), gr.Group(visible=True)
            conversion_type_selector.change(fn=switch_converter, inputs=conversion_type_selector, outputs=[ubfc_converter_group, scamps_converter_group])
            convert_button_ubfc.click(fn=convert_ubfc_to_mmpd, inputs=[ubfc_conv_vid, ubfc_conv_gt, ubfc_resize_option], outputs=[download_button_conv, status_text_conv])
            convert_button_scamps.click(fn=convert_scamps_to_mmpd, inputs=[scamps_conv_mat, scamps_resize_option], outputs=[download_button_conv, status_text_conv])
    
    gr.Markdown("---")
    gr.Markdown("This program was made by Indra Dewaji for the purpose of research in the Doctor of Professional Practice at Universiti Teknologi Malaysia")

demo.queue()
demo.launch(
    server_name="0.0.0.0",
    server_port=int(os.environ.get("PORT", 7860)),
    show_error=True,
    ssr_mode=False,
)