app.py

import json
import os

import cv2
import gradio as gr
import imagehash
import numpy as np
import plotly.graph_objects as go
from PIL import Image
from scipy.stats import pearsonr
from skimage.metrics import mean_squared_error as mse_skimage
from skimage.metrics import peak_signal_noise_ratio as psnr_skimage
from skimage.metrics import structural_similarity as ssim


class FrameMetrics:
    """Class to compute and store frame-by-frame metrics"""

    def __init__(self):
        self.metrics = {}

    def compute_ssim(self, frame1, frame2):
        """Compute SSIM between two frames"""
        if frame1 is None or frame2 is None:
            return None

        try:
            # Convert to grayscale for SSIM computation
            gray1 = (
                cv2.cvtColor(frame1, cv2.COLOR_RGB2GRAY)
                if len(frame1.shape) == 3
                else frame1
            )
            gray2 = (
                cv2.cvtColor(frame2, cv2.COLOR_RGB2GRAY)
                if len(frame2.shape) == 3
                else frame2
            )

            # Ensure both frames have the same dimensions
            if gray1.shape != gray2.shape:
                # Resize to match the smaller dimension
                h = min(gray1.shape[0], gray2.shape[0])
                w = min(gray1.shape[1], gray2.shape[1])
                gray1 = cv2.resize(gray1, (w, h))
                gray2 = cv2.resize(gray2, (w, h))

            # Compute SSIM
            ssim_value = ssim(gray1, gray2, data_range=255)
            return ssim_value

        except Exception as e:
            print(f"SSIM computation failed: {e}")
            return None

    def compute_ms_ssim(self, frame1, frame2):
        """Compute Multi-Scale SSIM between two frames"""
        if frame1 is None or frame2 is None:
            return None

        try:
            # Convert to grayscale for MS-SSIM computation
            gray1 = (
                cv2.cvtColor(frame1, cv2.COLOR_RGB2GRAY)
                if len(frame1.shape) == 3
                else frame1
            )
            gray2 = (
                cv2.cvtColor(frame2, cv2.COLOR_RGB2GRAY)
                if len(frame2.shape) == 3
                else frame2
            )

            # Ensure both frames have the same dimensions
            if gray1.shape != gray2.shape:
                h = min(gray1.shape[0], gray2.shape[0])
                w = min(gray1.shape[1], gray2.shape[1])
                gray1 = cv2.resize(gray1, (w, h))
                gray2 = cv2.resize(gray2, (w, h))

            # Ensure minimum size for multi-scale analysis
            min_size = 32
            if min(gray1.shape) < min_size:
                return None

            # Compute MS-SSIM using multiple scales
            from skimage.metrics import structural_similarity

            # Use win_size that works with image dimensions
            win_size = min(7, min(gray1.shape) // 4)
            if win_size < 3:
                win_size = 3

            ms_ssim_val = structural_similarity(
                gray1, gray2, data_range=255, win_size=win_size, multichannel=False
            )

            return ms_ssim_val

        except Exception as e:
            print(f"MS-SSIM computation failed: {e}")
            return None

    def compute_psnr(self, frame1, frame2):
        """Compute PSNR between two frames"""
        if frame1 is None or frame2 is None:
            return None

        try:
            # Ensure both frames have the same dimensions
            if frame1.shape != frame2.shape:
                h = min(frame1.shape[0], frame2.shape[0])
                w = min(frame1.shape[1], frame2.shape[1])
                c = (
                    min(frame1.shape[2], frame2.shape[2])
                    if len(frame1.shape) == 3
                    else 1
                )

                if len(frame1.shape) == 3:
                    frame1 = cv2.resize(frame1, (w, h))[:, :, :c]
                    frame2 = cv2.resize(frame2, (w, h))[:, :, :c]
                else:
                    frame1 = cv2.resize(frame1, (w, h))
                    frame2 = cv2.resize(frame2, (w, h))

            # Compute PSNR
            return psnr_skimage(frame1, frame2, data_range=255)
        except Exception as e:
            print(f"PSNR computation failed: {e}")
            return None

    def compute_mse(self, frame1, frame2):
        """Compute MSE between two frames"""
        if frame1 is None or frame2 is None:
            return None

        try:
            # Ensure both frames have the same dimensions
            if frame1.shape != frame2.shape:
                h = min(frame1.shape[0], frame2.shape[0])
                w = min(frame1.shape[1], frame2.shape[1])
                c = (
                    min(frame1.shape[2], frame2.shape[2])
                    if len(frame1.shape) == 3
                    else 1
                )

                if len(frame1.shape) == 3:
                    frame1 = cv2.resize(frame1, (w, h))[:, :, :c]
                    frame2 = cv2.resize(frame2, (w, h))[:, :, :c]
                else:
                    frame1 = cv2.resize(frame1, (w, h))
                    frame2 = cv2.resize(frame2, (w, h))

            # Compute MSE
            return mse_skimage(frame1, frame2)
        except Exception as e:
            print(f"MSE computation failed: {e}")
            return None

    def compute_phash(self, frame1, frame2):
        """Compute perceptual hash similarity between two frames"""
        if frame1 is None or frame2 is None:
            return None

        try:
            # Convert to PIL Images for imagehash
            pil1 = Image.fromarray(frame1)
            pil2 = Image.fromarray(frame2)

            # Compute perceptual hashes
            hash1 = imagehash.phash(pil1)
            hash2 = imagehash.phash(pil2)

            # Calculate similarity (lower hamming distance = more similar)
            hamming_distance = hash1 - hash2
            # Convert to similarity score (0-1, where 1 is identical)
            max_distance = len(str(hash1)) * 4  # 4 bits per hex char
            similarity = 1 - (hamming_distance / max_distance)

            return similarity
        except Exception as e:
            print(f"pHash computation failed: {e}")
            return None

    def compute_color_histogram_correlation(self, frame1, frame2):
        """Compute color histogram correlation between two frames"""
        if frame1 is None or frame2 is None:
            return None

        try:
            # Ensure both frames have the same dimensions
            if frame1.shape != frame2.shape:
                h = min(frame1.shape[0], frame2.shape[0])
                w = min(frame1.shape[1], frame2.shape[1])
                frame1 = cv2.resize(frame1, (w, h))
                frame2 = cv2.resize(frame2, (w, h))

            # Compute histograms for each channel
            correlations = []

            if len(frame1.shape) == 3:  # Color image
                for i in range(3):  # R, G, B channels
                    hist1 = cv2.calcHist([frame1], [i], None, [256], [0, 256])
                    hist2 = cv2.calcHist([frame2], [i], None, [256], [0, 256])

                    # Flatten histograms
                    hist1 = hist1.flatten()
                    hist2 = hist2.flatten()

                    # Compute correlation
                    if np.std(hist1) > 0 and np.std(hist2) > 0:
                        corr, _ = pearsonr(hist1, hist2)
                        correlations.append(corr)

                # Return average correlation across channels
                return np.mean(correlations) if correlations else 0.0
            else:  # Grayscale
                hist1 = cv2.calcHist([frame1], [0], None, [256], [0, 256]).flatten()
                hist2 = cv2.calcHist([frame2], [0], None, [256], [0, 256]).flatten()

                if np.std(hist1) > 0 and np.std(hist2) > 0:
                    corr, _ = pearsonr(hist1, hist2)
                    return corr
                else:
                    return 0.0

        except Exception as e:
            print(f"Color histogram correlation computation failed: {e}")
            return None

    def compute_sharpness(self, frame):
        """Compute sharpness using Laplacian variance method"""
        if frame is None:
            return None

        # Convert to grayscale if needed
        gray = (
            cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY) if len(frame.shape) == 3 else frame
        )

        # Compute Laplacian variance (higher values = sharper)
        laplacian = cv2.Laplacian(gray, cv2.CV_64F)
        sharpness = laplacian.var()

        return sharpness

    def compute_frame_metrics(self, frame1, frame2, frame_idx):
        """Compute all metrics for a frame pair"""
        metrics = {
            "frame_index": frame_idx,
            "ssim": self.compute_ssim(frame1, frame2),
            "psnr": self.compute_psnr(frame1, frame2),
            "mse": self.compute_mse(frame1, frame2),
            "phash": self.compute_phash(frame1, frame2),
            "color_hist_corr": self.compute_color_histogram_correlation(frame1, frame2),
            "sharpness1": self.compute_sharpness(frame1),
            "sharpness2": self.compute_sharpness(frame2),
        }

        # Compute average sharpness for the pair
        if metrics["sharpness1"] is not None and metrics["sharpness2"] is not None:
            metrics["sharpness_avg"] = (
                metrics["sharpness1"] + metrics["sharpness2"]
            ) / 2
            metrics["sharpness_diff"] = abs(
                metrics["sharpness1"] - metrics["sharpness2"]
            )
        else:
            metrics["sharpness_avg"] = None
            metrics["sharpness_diff"] = None

        return metrics

    def compute_all_metrics(self, frames1, frames2):
        """Compute metrics for all frame pairs"""
        all_metrics = []
        max_frames = max(len(frames1), len(frames2))

        for i in range(max_frames):
            frame1 = frames1[i] if i < len(frames1) else None
            frame2 = frames2[i] if i < len(frames2) else None

            if frame1 is not None or frame2 is not None:
                metrics = self.compute_frame_metrics(frame1, frame2, i)
                all_metrics.append(metrics)
            else:
                # Handle cases where both frames are missing
                all_metrics.append(
                    {
                        "frame_index": i,
                        "ssim": None,
                        "ms_ssim": None,
                        "psnr": None,
                        "mse": None,
                        "phash": None,
                        "color_hist_corr": None,
                        "sharpness1": None,
                        "sharpness2": None,
                        "sharpness_avg": None,
                        "sharpness_diff": None,
                    }
                )

        return all_metrics

    def get_metric_summary(self, metrics_list):
        """Compute summary statistics for all metrics"""
        metric_names = [
            "ssim",
            "psnr",
            "mse",
            "phash",
            "color_hist_corr",
            "sharpness1",
            "sharpness2",
            "sharpness_avg",
            "sharpness_diff",
        ]

        summary = {
            "total_frames": len(metrics_list),
            "valid_frames": len([m for m in metrics_list if m.get("ssim") is not None]),
        }

        # Compute statistics for each metric
        for metric_name in metric_names:
            valid_values = [
                m[metric_name] for m in metrics_list if m.get(metric_name) is not None
            ]

            if valid_values:
                summary.update(
                    {
                        f"{metric_name}_mean": np.mean(valid_values),
                        f"{metric_name}_min": np.min(valid_values),
                        f"{metric_name}_max": np.max(valid_values),
                        f"{metric_name}_std": np.std(valid_values),
                    }
                )

        return summary

    def create_individual_metric_plots(self, metrics_list, current_frame=0):
        """Create individual plots for each metric with frame on x-axis"""
        if not metrics_list:
            return None

        # Extract frame indices
        frame_indices = [m["frame_index"] for m in metrics_list]

        # Helper function to get valid data
        def get_valid_data(metric_name):
            values = [m.get(metric_name) for m in metrics_list]
            valid_indices = [i for i, v in enumerate(values) if v is not None]
            valid_values = [values[i] for i in valid_indices]
            valid_frames = [frame_indices[i] for i in valid_indices]
            return valid_frames, valid_values

        # Create individual plots for each metric
        plots = {}

        # 1. SSIM Plot
        ssim_frames, ssim_values = get_valid_data("ssim")
        if ssim_values:
            fig_ssim = go.Figure()
            fig_ssim.add_trace(
                go.Scatter(
                    x=ssim_frames,
                    y=ssim_values,
                    mode="lines+markers",
                    name="SSIM",
                    line=dict(color="blue", width=3),
                    marker=dict(size=6),
                    hovertemplate="<b>Frame %{x}</b><br>SSIM: %{y:.4f}<extra></extra>",
                )
            )

            if current_frame is not None:
                fig_ssim.add_vline(
                    x=current_frame,
                    line_dash="dash",
                    line_color="red",
                    line_width=2,
                )

            fig_ssim.update_layout(
                height=300,
                margin=dict(t=20, b=40, l=60, r=20),
                plot_bgcolor="rgba(0,0,0,0)",
                paper_bgcolor="rgba(0,0,0,0)",
                showlegend=False,
            )
            fig_ssim.update_xaxes(title_text="Frame")
            fig_ssim.update_yaxes(title_text="SSIM", range=[0, 1.05])
            plots["ssim"] = fig_ssim

        # 2. PSNR Plot
        psnr_frames, psnr_values = get_valid_data("psnr")
        if psnr_values:
            fig_psnr = go.Figure()
            fig_psnr.add_trace(
                go.Scatter(
                    x=psnr_frames,
                    y=psnr_values,
                    mode="lines+markers",
                    name="PSNR",
                    line=dict(color="green", width=3),
                    marker=dict(size=6),
                    hovertemplate="<b>Frame %{x}</b><br>PSNR: %{y:.2f} dB<extra></extra>",
                )
            )

            if current_frame is not None:
                fig_psnr.add_vline(
                    x=current_frame,
                    line_dash="dash",
                    line_color="red",
                    line_width=2,
                )

            fig_psnr.update_layout(
                height=300,
                margin=dict(t=20, b=40, l=60, r=20),
                plot_bgcolor="rgba(0,0,0,0)",
                paper_bgcolor="rgba(0,0,0,0)",
                showlegend=False,
            )
            fig_psnr.update_xaxes(title_text="Frame")
            fig_psnr.update_yaxes(title_text="PSNR (dB)")
            plots["psnr"] = fig_psnr

        # 3. MSE Plot
        mse_frames, mse_values = get_valid_data("mse")
        if mse_values:
            fig_mse = go.Figure()
            fig_mse.add_trace(
                go.Scatter(
                    x=mse_frames,
                    y=mse_values,
                    mode="lines+markers",
                    name="MSE",
                    line=dict(color="red", width=3),
                    marker=dict(size=6),
                    hovertemplate="<b>Frame %{x}</b><br>MSE: %{y:.2f}<extra></extra>",
                )
            )

            if current_frame is not None:
                fig_mse.add_vline(
                    x=current_frame,
                    line_dash="dash",
                    line_color="red",
                    line_width=2,
                )

            fig_mse.update_layout(
                height=300,
                margin=dict(t=20, b=40, l=60, r=20),
                plot_bgcolor="rgba(0,0,0,0)",
                paper_bgcolor="rgba(0,0,0,0)",
                showlegend=False,
            )
            fig_mse.update_xaxes(title_text="Frame")
            fig_mse.update_yaxes(title_text="MSE")
            plots["mse"] = fig_mse

        # 4. pHash Plot
        phash_frames, phash_values = get_valid_data("phash")
        if phash_values:
            fig_phash = go.Figure()
            fig_phash.add_trace(
                go.Scatter(
                    x=phash_frames,
                    y=phash_values,
                    mode="lines+markers",
                    name="pHash",
                    line=dict(color="purple", width=3),
                    marker=dict(size=6),
                    hovertemplate="<b>Frame %{x}</b><br>pHash: %{y:.4f}<extra></extra>",
                )
            )

            if current_frame is not None:
                fig_phash.add_vline(
                    x=current_frame,
                    line_dash="dash",
                    line_color="red",
                    line_width=2,
                )

            fig_phash.update_layout(
                height=300,
                margin=dict(t=20, b=40, l=60, r=20),
                plot_bgcolor="rgba(0,0,0,0)",
                paper_bgcolor="rgba(0,0,0,0)",
                showlegend=False,
            )
            fig_phash.update_xaxes(title_text="Frame")
            fig_phash.update_yaxes(title_text="pHash Similarity")
            plots["phash"] = fig_phash

        # 5. Color Histogram Correlation Plot
        hist_frames, hist_values = get_valid_data("color_hist_corr")
        if hist_values:
            fig_hist = go.Figure()
            fig_hist.add_trace(
                go.Scatter(
                    x=hist_frames,
                    y=hist_values,
                    mode="lines+markers",
                    name="Color Histogram",
                    line=dict(color="orange", width=3),
                    marker=dict(size=6),
                    hovertemplate="<b>Frame %{x}</b><br>Color Histogram: %{y:.4f}<extra></extra>",
                )
            )

            if current_frame is not None:
                fig_hist.add_vline(
                    x=current_frame,
                    line_dash="dash",
                    line_color="red",
                    line_width=2,
                )

            fig_hist.update_layout(
                height=300,
                margin=dict(t=20, b=40, l=60, r=20),
                plot_bgcolor="rgba(0,0,0,0)",
                paper_bgcolor="rgba(0,0,0,0)",
                showlegend=False,
            )
            fig_hist.update_xaxes(title_text="Frame")
            fig_hist.update_yaxes(title_text="Color Histogram Correlation")
            plots["color_hist"] = fig_hist

        # 6. Sharpness Comparison Plot
        sharp1_frames, sharp1_values = get_valid_data("sharpness1")
        sharp2_frames, sharp2_values = get_valid_data("sharpness2")

        if sharp1_values or sharp2_values:
            fig_sharp = go.Figure()

            if sharp1_values:
                fig_sharp.add_trace(
                    go.Scatter(
                        x=sharp1_frames,
                        y=sharp1_values,
                        mode="lines+markers",
                        name="Video 1",
                        line=dict(color="darkgreen", width=3),
                        marker=dict(size=6),
                        hovertemplate="<b>Frame %{x}</b><br>Video 1 Sharpness: %{y:.1f}<extra></extra>",
                    )
                )

            if sharp2_values:
                fig_sharp.add_trace(
                    go.Scatter(
                        x=sharp2_frames,
                        y=sharp2_values,
                        mode="lines+markers",
                        name="Video 2",
                        line=dict(color="darkblue", width=3),
                        marker=dict(size=6),
                        hovertemplate="<b>Frame %{x}</b><br>Video 2 Sharpness: %{y:.1f}<extra></extra>",
                    )
                )

            if current_frame is not None:
                fig_sharp.add_vline(
                    x=current_frame,
                    line_dash="dash",
                    line_color="red",
                    line_width=2,
                )

            fig_sharp.update_layout(
                height=300,
                margin=dict(t=20, b=40, l=60, r=20),
                plot_bgcolor="rgba(0,0,0,0)",
                paper_bgcolor="rgba(0,0,0,0)",
                showlegend=True,
                legend=dict(
                    orientation="h", yanchor="bottom", y=1.02, xanchor="center", x=0.5
                ),
            )
            fig_sharp.update_xaxes(title_text="Frame")
            fig_sharp.update_yaxes(title_text="Sharpness")
            plots["sharpness"] = fig_sharp

        # 7. Overall Quality Score Plot (Combination of metrics)
        # Calculate overall quality score by combining normalized metrics
        if ssim_values and psnr_values and len(ssim_values) == len(psnr_values):
            # Get data for metrics that contribute to overall score
            phash_frames_overall, phash_values_overall = get_valid_data("phash")

            # Ensure we have the same frames for all metrics
            common_frames = set(ssim_frames) & set(psnr_frames)
            if phash_values_overall:
                common_frames = common_frames & set(phash_frames_overall)

            common_frames = sorted(list(common_frames))

            if common_frames:
                # Extract values for common frames
                ssim_common = [
                    ssim_values[ssim_frames.index(f)]
                    for f in common_frames
                    if f in ssim_frames
                ]
                psnr_common = [
                    psnr_values[psnr_frames.index(f)]
                    for f in common_frames
                    if f in psnr_frames
                ]

                # Normalize PSNR to 0-1 scale (typical range 0-50dB)
                psnr_normalized = [min(p / 50.0, 1.0) for p in psnr_common]

                # Start with SSIM and normalized PSNR
                quality_components = [ssim_common, psnr_normalized]
                component_names = ["SSIM", "PSNR"]

                # Add pHash if available
                if phash_values_overall:
                    phash_common = [
                        phash_values_overall[phash_frames_overall.index(f)]
                        for f in common_frames
                        if f in phash_frames_overall
                    ]
                    if len(phash_common) == len(ssim_common):
                        quality_components.append(phash_common)
                        component_names.append("pHash")

                # Calculate average across all components
                overall_quality = []
                for i in range(len(common_frames)):
                    frame_scores = [
                        component[i]
                        for component in quality_components
                        if i < len(component)
                    ]
                    overall_quality.append(sum(frame_scores) / len(frame_scores))

                fig_overall = go.Figure()
                fig_overall.add_trace(
                    go.Scatter(
                        x=common_frames,
                        y=overall_quality,
                        mode="lines+markers",
                        name="Overall Quality",
                        line=dict(color="gold", width=4),
                        marker=dict(size=8),
                        hovertemplate="<b>Frame %{x}</b><br>Overall Quality: %{y:.3f}<br><i>Combined from: "
                        + ", ".join(component_names)
                        + "</i><extra></extra>",
                        fill="tonexty",
                    )
                )

                if current_frame is not None:
                    fig_overall.add_vline(
                        x=current_frame,
                        line_dash="dash",
                        line_color="red",
                        line_width=2,
                    )

                fig_overall.update_layout(
                    height=300,
                    margin=dict(t=20, b=40, l=60, r=20),
                    plot_bgcolor="rgba(0,0,0,0)",
                    paper_bgcolor="rgba(0,0,0,0)",
                    showlegend=False,
                )
                fig_overall.update_xaxes(title_text="Frame")
                fig_overall.update_yaxes(
                    title_text="Overall Quality Score", range=[0, 1.05]
                )
                plots["overall"] = fig_overall

        return plots

    def create_modern_plot(self, metrics_list, current_frame=0):
        """Create individual metric plots instead of combined dashboard"""
        return self.create_individual_metric_plots(metrics_list, current_frame)


class VideoFrameComparator:
    def __init__(self):
        self.video1_frames = []
        self.video2_frames = []
        self.max_frames = 0
        self.frame_metrics = FrameMetrics()
        self.computed_metrics = []
        self.metrics_summary = {}

    def extract_frames(self, video_path):
        """Extract all frames from a video file or URL"""
        if not video_path:
            return []

        # Check if it's a URL or local file
        is_url = video_path.startswith(("http://", "https://"))

        if not is_url and not os.path.exists(video_path):
            print(f"Warning: Local video file not found: {video_path}")
            return []

        frames = []
        cap = cv2.VideoCapture(video_path)

        if not cap.isOpened():
            print(
                f"Error: Could not open video {'URL' if is_url else 'file'}: {video_path}"
            )
            return []

        try:
            frame_count = 0
            while True:
                ret, frame = cap.read()
                if not ret:
                    break
                # Convert BGR to RGB for display
                frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
                frames.append(frame_rgb)
                frame_count += 1

                # Add progress feedback for URLs (which might be slower)
                if is_url and frame_count % 30 == 0:
                    print(f"Processed {frame_count} frames from URL...")

        except Exception as e:
            print(f"Error processing video: {e}")
        finally:
            cap.release()

        print(
            f"Successfully extracted {len(frames)} frames from {'URL' if is_url else 'file'}: {video_path}"
        )
        return frames

    def is_comparison_in_data_json(
        self, video1_path, video2_path, json_file_path="data.json"
    ):
        """Check if this video comparison exists in data.json"""
        try:
            with open(json_file_path, "r") as f:
                data = json.load(f)

            for comparison in data.get("comparisons", []):
                videos = comparison.get("videos", [])
                if len(videos) == 2:
                    # Check both orders (works for both local files and URLs)
                    if (videos[0] == video1_path and videos[1] == video2_path) or (
                        videos[0] == video2_path and videos[1] == video1_path
                    ):
                        return True

            return False
        except:
            return False

    def load_videos(self, video1_path, video2_path):
        """Load both videos and extract frames"""
        if not video1_path and not video2_path:
            return "Please upload at least one video.", 0, None, None, "", None

        # Extract frames from both videos
        self.video1_frames = self.extract_frames(video1_path) if video1_path else []
        self.video2_frames = self.extract_frames(video2_path) if video2_path else []

        # Determine maximum number of frames
        self.max_frames = max(len(self.video1_frames), len(self.video2_frames))

        if self.max_frames == 0:
            return (
                "No valid frames found in the uploaded videos.",
                0,
                None,
                None,
                "",
                None,
            )

        # Compute metrics if both videos are present and not in data.json
        metrics_info = ""
        plots = None

        if (
            video1_path
            and video2_path
            and not self.is_comparison_in_data_json(video1_path, video2_path)
        ):
            print("Computing comprehensive frame-by-frame metrics...")
            self.computed_metrics = self.frame_metrics.compute_all_metrics(
                self.video1_frames, self.video2_frames
            )
            self.metrics_summary = self.frame_metrics.get_metric_summary(
                self.computed_metrics
            )

            # Build metrics info string
            metrics_info = "\n\n📊 Computed Metrics Summary:\n"

            metric_display = {
                "ssim": ("SSIM", ".4f", "", "↑ Higher=Better"),
                "psnr": ("PSNR", ".2f", " dB", "↑ Higher=Better"),
                "mse": ("MSE", ".2f", "", "↓ Lower=Better"),
                "phash": ("pHash", ".4f", "", "↑ Higher=Better"),
                "color_hist_corr": ("Color Hist", ".4f", "", "↑ Higher=Better"),
                "sharpness_avg": ("Sharpness", ".1f", "", "↑ Higher=Better"),
            }

            for metric_key, (
                display_name,
                format_str,
                unit,
                direction,
            ) in metric_display.items():
                if self.metrics_summary.get(f"{metric_key}_mean") is not None:
                    mean_val = self.metrics_summary[f"{metric_key}_mean"]
                    std_val = self.metrics_summary[f"{metric_key}_std"]
                    metrics_info += f"{display_name}: μ={mean_val:{format_str}}{unit}, σ={std_val:{format_str}}{unit} ({direction})\n"

            metrics_info += f"Valid Frames: {self.metrics_summary['valid_frames']}/{self.metrics_summary['total_frames']}"

            # Generate initial plot
            plots = self.frame_metrics.create_individual_metric_plots(
                self.computed_metrics, 0
            )
        else:
            self.computed_metrics = []
            self.metrics_summary = {}
            if video1_path and video2_path:
                metrics_info = "\n\n📋 Note: This comparison is predefined in data.json (metrics not computed)"

        # Get initial frames
        frame1 = (
            self.video1_frames[0]
            if self.video1_frames
            else np.zeros((480, 640, 3), dtype=np.uint8)
        )
        frame2 = (
            self.video2_frames[0]
            if self.video2_frames
            else np.zeros((480, 640, 3), dtype=np.uint8)
        )

        status_msg = "Videos loaded successfully!\n"
        status_msg += f"Video 1: {len(self.video1_frames)} frames\n"
        status_msg += f"Video 2: {len(self.video2_frames)} frames\n"
        status_msg += (
            f"Use the slider to navigate through frames (0-{self.max_frames - 1})"
        )
        status_msg += metrics_info

        return (
            status_msg,
            self.max_frames - 1,
            frame1,
            frame2,
            self.get_current_frame_info(0),
            plots,
        )

    def get_frames_at_index(self, frame_index):
        """Get frames at specific index from both videos"""
        frame_index = int(frame_index)

        # Get frame from video 1
        if frame_index < len(self.video1_frames):
            frame1 = self.video1_frames[frame_index]
        else:
            # Create a placeholder if frame doesn't exist
            frame1 = np.zeros((480, 640, 3), dtype=np.uint8)
            cv2.putText(
                frame1,
                f"Frame {frame_index} not available",
                (50, 240),
                cv2.FONT_HERSHEY_SIMPLEX,
                1,
                (255, 255, 255),
                2,
            )

        # Get frame from video 2
        if frame_index < len(self.video2_frames):
            frame2 = self.video2_frames[frame_index]
        else:
            # Create a placeholder if frame doesn't exist
            frame2 = np.zeros((480, 640, 3), dtype=np.uint8)
            cv2.putText(
                frame2,
                f"Frame {frame_index} not available",
                (50, 240),
                cv2.FONT_HERSHEY_SIMPLEX,
                1,
                (255, 255, 255),
                2,
            )

        return frame1, frame2

    def get_current_frame_info(self, frame_index):
        """Get information about the current frame including metrics"""
        frame_index = int(frame_index)
        info = f"Current Frame: {frame_index} / {self.max_frames - 1}"

        # Add metrics info if available
        if self.computed_metrics and frame_index < len(self.computed_metrics):
            metrics = self.computed_metrics[frame_index]

            # === COMPARISON METRICS (Between Videos) ===
            comparison_metrics = []

            # SSIM with quality assessment
            if metrics.get("ssim") is not None:
                ssim_val = metrics["ssim"]
                if ssim_val >= 0.9:
                    quality = "🟢 Excellent"
                elif ssim_val >= 0.8:
                    quality = "🔵 Good"
                elif ssim_val >= 0.6:
                    quality = "🟡 Fair"
                else:
                    quality = "🔴 Poor"
                comparison_metrics.append(
                    f"SSIM: {ssim_val:.4f} ({quality} similarity)"
                )

            # PSNR with quality indicator
            if metrics.get("psnr") is not None:
                psnr_val = metrics["psnr"]
                if psnr_val >= 40:
                    psnr_quality = "🟢 Excellent"
                elif psnr_val >= 30:
                    psnr_quality = "🔵 Good"
                elif psnr_val >= 20:
                    psnr_quality = "🟡 Fair"
                else:
                    psnr_quality = "🔴 Poor"
                comparison_metrics.append(
                    f"PSNR: {psnr_val:.1f}dB ({psnr_quality} signal quality)"
                )

            # MSE with quality indicator (lower is better)
            if metrics.get("mse") is not None:
                mse_val = metrics["mse"]
                if mse_val <= 50:
                    mse_quality = "🟢 Very Similar"
                elif mse_val <= 100:
                    mse_quality = "🔵 Similar"
                elif mse_val <= 200:
                    mse_quality = "🟡 Moderately Different"
                else:
                    mse_quality = "🔴 Very Different"
                comparison_metrics.append(f"MSE: {mse_val:.1f} ({mse_quality})")

            # pHash with quality indicator
            if metrics.get("phash") is not None:
                phash_val = metrics["phash"]
                if phash_val >= 0.95:
                    phash_quality = "🟢 Nearly Identical"
                elif phash_val >= 0.9:
                    phash_quality = "🔵 Very Similar"
                elif phash_val >= 0.8:
                    phash_quality = "🟡 Somewhat Similar"
                else:
                    phash_quality = "🔴 Different"
                comparison_metrics.append(
                    f"pHash: {phash_val:.3f} ({phash_quality} perceptually)"
                )

            # Color Histogram Correlation
            if metrics.get("color_hist_corr") is not None:
                color_val = metrics["color_hist_corr"]
                if color_val >= 0.9:
                    color_quality = "🟢 Very Similar Colors"
                elif color_val >= 0.8:
                    color_quality = "🔵 Similar Colors"
                elif color_val >= 0.6:
                    color_quality = "🟡 Moderate Color Diff"
                else:
                    color_quality = "🔴 Different Colors"
                comparison_metrics.append(f"Color: {color_val:.3f} ({color_quality})")

            # Add comparison metrics to info
            if comparison_metrics:
                info += "\n📊 Comparison Analysis: " + " | ".join(comparison_metrics)

            # === INDIVIDUAL VIDEO QUALITY ===
            individual_metrics = []

            # Individual Sharpness for each video
            if metrics.get("sharpness1") is not None:
                sharp1 = metrics["sharpness1"]
                if sharp1 >= 200:
                    sharp1_quality = "🟢 Sharp"
                elif sharp1 >= 100:
                    sharp1_quality = "🔵 Moderate"
                elif sharp1 >= 50:
                    sharp1_quality = "🟡 Soft"
                else:
                    sharp1_quality = "🔴 Blurry"
                individual_metrics.append(
                    f"V1 Sharpness: {sharp1:.0f} ({sharp1_quality})"
                )

            if metrics.get("sharpness2") is not None:
                sharp2 = metrics["sharpness2"]
                if sharp2 >= 200:
                    sharp2_quality = "🟢 Sharp"
                elif sharp2 >= 100:
                    sharp2_quality = "🔵 Moderate"
                elif sharp2 >= 50:
                    sharp2_quality = "🟡 Soft"
                else:
                    sharp2_quality = "🔴 Blurry"
                individual_metrics.append(
                    f"V2 Sharpness: {sharp2:.0f} ({sharp2_quality})"
                )

            # Sharpness comparison
            if (
                metrics.get("sharpness1") is not None
                and metrics.get("sharpness2") is not None
            ):
                sharp1 = metrics["sharpness1"]
                sharp2 = metrics["sharpness2"]

                # Calculate difference percentage
                diff_pct = abs(sharp1 - sharp2) / max(sharp1, sharp2) * 100

                # Determine significance with clearer labels
                if diff_pct > 20:
                    significance = "🔴 MAJOR difference"
                elif diff_pct > 10:
                    significance = "🟡 MODERATE difference"
                elif diff_pct > 5:
                    significance = "🔵 MINOR difference"
                else:
                    significance = "🟢 NEGLIGIBLE difference"

                # Determine which is sharper
                if sharp1 > sharp2:
                    comparison = "V1 is sharper"
                elif sharp2 > sharp1:
                    comparison = "V2 is sharper"
                else:
                    comparison = "Equal sharpness"

                individual_metrics.append(f"Sharpness: {comparison} ({significance})")

            # Add individual metrics to info
            if individual_metrics:
                info += "\n🎯 Individual Quality: " + " | ".join(individual_metrics)

            # === OVERALL QUALITY ASSESSMENT ===
            # Calculate combined quality score from multiple metrics
            quality_score = 0
            quality_count = 0
            metric_contributions = []

            # SSIM contribution
            if metrics.get("ssim") is not None:
                quality_score += metrics["ssim"]
                quality_count += 1
                metric_contributions.append(f"SSIM({metrics['ssim']:.3f})")

            # PSNR contribution (normalized to 0-1 scale)
            if metrics.get("psnr") is not None:
                psnr_norm = min(metrics["psnr"] / 50, 1.0)
                quality_score += psnr_norm
                quality_count += 1
                metric_contributions.append(f"PSNR({psnr_norm:.3f})")

            # pHash contribution
            if metrics.get("phash") is not None:
                quality_score += metrics["phash"]
                quality_count += 1
                metric_contributions.append(f"pHash({metrics['phash']:.3f})")

            if quality_count > 0:
                avg_quality = quality_score / quality_count

                # Add overall assessment with formula explanation
                if avg_quality >= 0.9:
                    overall = "✨ Excellent Overall"
                elif avg_quality >= 0.8:
                    overall = "✅ Good Overall"
                elif avg_quality >= 0.6:
                    overall = "⚠️ Fair Overall"
                else:
                    overall = "❌ Poor Overall"

                info += f"\n🎯 Overall Quality: {avg_quality:.3f} ({overall})"
                info += f"\n   Formula: Average of {' + '.join(metric_contributions)} = {avg_quality:.3f}"

        return info

    def get_updated_plot(self, frame_index):
        """Get updated plot with current frame highlighted"""
        if self.computed_metrics:
            return self.frame_metrics.create_individual_metric_plots(
                self.computed_metrics, int(frame_index)
            )
        return None


def load_examples_from_json(json_file_path="data.json"):
    """Load example video pairs from JSON configuration file"""
    try:
        with open(json_file_path, "r") as f:
            data = json.load(f)

        examples = []

        # Extract video pairs from the comparisons
        for comparison in data.get("comparisons", []):
            videos = comparison.get("videos", [])

            # Validate that video files/URLs exist or are accessible
            valid_videos = []
            for video_path in videos:
                if video_path:  # Check if not empty/None
                    # Check if it's a URL
                    if video_path.startswith(("http://", "https://")):
                        # For URLs, we'll assume they're valid (can't easily check without downloading)
                        # OpenCV will handle the validation during actual loading
                        valid_videos.append(video_path)
                        print(f"Added video URL: {video_path}")
                    elif os.path.exists(video_path):
                        # For local files, check existence
                        valid_videos.append(video_path)
                        print(f"Added local video file: {video_path}")
                    else:
                        print(f"Warning: Local video file not found: {video_path}")

            # Add to examples if we have valid videos
            if len(valid_videos) == 2:
                examples.append(valid_videos)
            elif len(valid_videos) == 1:
                # Single video example (compare with None)
                examples.append([valid_videos[0], None])

        return examples

    except FileNotFoundError:
        print(f"Warning: {json_file_path} not found. No examples will be loaded.")
        return []
    except json.JSONDecodeError as e:
        print(f"Error parsing {json_file_path}: {e}")
        return []
    except Exception as e:
        print(f"Error loading examples: {e}")
        return []


def get_all_videos_from_json(json_file_path="data.json"):
    """Get list of all unique videos mentioned in the JSON file"""
    try:
        with open(json_file_path, "r") as f:
            data = json.load(f)

        all_videos = set()

        # Extract all unique video paths/URLs from comparisons
        for comparison in data.get("comparisons", []):
            videos = comparison.get("videos", [])
            for video_path in videos:
                if video_path:  # Only add non-empty paths
                    # Check if it's a URL or local file
                    if video_path.startswith(("http://", "https://")):
                        # For URLs, add them directly
                        all_videos.add(video_path)
                    elif os.path.exists(video_path):
                        # For local files, check existence before adding
                        all_videos.add(video_path)

        return sorted(list(all_videos))

    except FileNotFoundError:
        print(f"Warning: {json_file_path} not found.")
        return []
    except json.JSONDecodeError as e:
        print(f"Error parsing {json_file_path}: {e}")
        return []
    except Exception as e:
        print(f"Error loading videos: {e}")
        return []


def create_app():
    comparator = VideoFrameComparator()
    example_pairs = load_examples_from_json()
    all_videos = get_all_videos_from_json()

    with gr.Blocks(
        title="Frame Arena - Video Frame Comparator",
        # theme=gr.themes.Soft(),
    ) as app:
        gr.Markdown("""
        # 🎬 Frame Arena: Frame by frame comparisons of any videos

        - Upload videos in common formats with the same number of frames (MP4, AVI, MOV, etc.) or use URLs
        - **7 Quality Metrics**: SSIM, PSNR, MSE, pHash, Color Histogram, Sharpness + Overall Quality
        - **Individual Visualization**: Each metric gets its own dedicated plot
        - **Real-time Analysis**: Navigate frames with live metric updates
        - **Smart Comparisons**: Understand differences between videos per metric

        **Perfect for**: Analyzing compression effects, processing artifacts, visual quality assessment, and compression algorithm comparisons.
        """)

        with gr.Row():
            with gr.Column():
                gr.Markdown("### Video 1")
                video1_input = gr.File(
                    label="Upload Video 1",
                    file_types=[
                        ".mp4",
                        ".avi",
                        ".mov",
                        ".mkv",
                        ".wmv",
                        ".flv",
                        ".webm",
                    ],
                    type="filepath",
                )

            with gr.Column():
                gr.Markdown("### Video 2")
                video2_input = gr.File(
                    label="Upload Video 2",
                    file_types=[
                        ".mp4",
                        ".avi",
                        ".mov",
                        ".mkv",
                        ".wmv",
                        ".flv",
                        ".webm",
                    ],
                    type="filepath",
                )

        # Add examples if available (this auto-populates inputs when clicked)
        if example_pairs:
            gr.Markdown("### 📁 Example Video Comparisons")
            gr.Examples(
                examples=example_pairs,
                inputs=[video1_input, video2_input],
                label="Click any example to load video pairs:",
                examples_per_page=10,
            )

        load_btn = gr.Button("🔄 Load Videos", variant="primary", size="lg")

        # Frame comparison section (initially hidden)
        frame_display = gr.Row(visible=False)
        with frame_display:
            with gr.Column():
                gr.Markdown("### Video 1 - Current Frame")
                frame1_output = gr.Image(
                    label="Video 1 Frame", type="numpy", interactive=False, height=400
                )

            with gr.Column():
                gr.Markdown("### Video 2 - Current Frame")
                frame2_output = gr.Image(
                    label="Video 2 Frame", type="numpy", interactive=False, height=400
                )

        # Frame navigation (initially hidden) - moved underneath frames
        frame_controls = gr.Row(visible=False)
        with frame_controls:
            frame_slider = gr.Slider(
                minimum=0,
                maximum=0,
                step=1,
                value=0,
                label="Frame Number",
                interactive=False,
            )

        # Comprehensive metrics visualization (initially hidden)
        metrics_section = gr.Row(visible=False)
        with metrics_section:
            with gr.Column():
                # Frame info moved above the plot
                frame_info = gr.Textbox(
                    label="Frame Information & Metrics",
                    interactive=False,
                    value="",
                    lines=3,
                )
                gr.Markdown("### 📊 Individual Metric Analysis")

                # Overall quality plot
                with gr.Row():
                    overall_plot = gr.Plot(
                        label="Overall Quality (Combined Metric [SSIM + normalized_PSNR + pHash])",
                        show_label=True,
                    )

                # Individual metric plots
                with gr.Row():
                    ssim_plot = gr.Plot(label="SSIM", show_label=True)
                    psnr_plot = gr.Plot(label="PSNR", show_label=True)

                with gr.Row():
                    mse_plot = gr.Plot(label="MSE", show_label=True)
                    phash_plot = gr.Plot(label="pHash", show_label=True)

                with gr.Row():
                    color_plot = gr.Plot(label="Color Histogram", show_label=True)
                    sharpness_plot = gr.Plot(label="Sharpness", show_label=True)

        # Status and frame info (moved below plots, initially hidden)
        info_section = gr.Row(visible=False)
        with info_section:
            with gr.Column():
                status_output = gr.Textbox(label="Status", interactive=False, lines=8)

        # Event handlers
        def load_videos_handler(video1, video2):
            status, max_frames, frame1, frame2, info, plots = comparator.load_videos(
                video1, video2
            )

            # Update slider
            slider_update = gr.Slider(
                minimum=0,
                maximum=max_frames,
                step=1,
                value=0,
                interactive=True if max_frames > 0 else False,
            )

            # Show/hide sections based on whether videos were loaded successfully
            videos_loaded = max_frames > 0

            # Extract individual plots from the plots dictionary
            ssim_fig = plots.get("ssim") if plots else None
            psnr_fig = plots.get("psnr") if plots else None
            mse_fig = plots.get("mse") if plots else None
            phash_fig = plots.get("phash") if plots else None
            color_fig = plots.get("color_hist") if plots else None
            sharpness_fig = plots.get("sharpness") if plots else None
            overall_fig = plots.get("overall") if plots else None

            return (
                status,  # status_output
                slider_update,  # frame_slider
                frame1,  # frame1_output
                frame2,  # frame2_output
                info,  # frame_info
                ssim_fig,  # ssim_plot
                psnr_fig,  # psnr_plot
                mse_fig,  # mse_plot
                phash_fig,  # phash_plot
                color_fig,  # color_plot
                sharpness_fig,  # sharpness_plot
                overall_fig,  # overall_plot
                gr.Row(visible=videos_loaded),  # frame_controls
                gr.Row(visible=videos_loaded),  # frame_display
                gr.Row(visible=videos_loaded),  # metrics_section
                gr.Row(visible=videos_loaded),  # info_section
            )

        def update_frames(frame_index):
            if comparator.max_frames == 0:
                return (
                    None,
                    None,
                    "No videos loaded",
                    None,
                    None,
                    None,
                    None,
                    None,
                    None,
                    None,
                )

            frame1, frame2 = comparator.get_frames_at_index(frame_index)
            info = comparator.get_current_frame_info(frame_index)
            plots = comparator.get_updated_plot(frame_index)

            # Extract individual plots from the plots dictionary
            ssim_fig = plots.get("ssim") if plots else None
            psnr_fig = plots.get("psnr") if plots else None
            mse_fig = plots.get("mse") if plots else None
            phash_fig = plots.get("phash") if plots else None
            color_fig = plots.get("color_hist") if plots else None
            sharpness_fig = plots.get("sharpness") if plots else None
            overall_fig = plots.get("overall") if plots else None

            return (
                frame1,
                frame2,
                info,
                ssim_fig,
                psnr_fig,
                mse_fig,
                phash_fig,
                color_fig,
                sharpness_fig,
                overall_fig,
            )

        # Auto-load when examples populate the inputs
        def auto_load_when_examples_change(video1, video2):
            # Only auto-load if both inputs are provided (from examples)
            if video1 and video2:
                return load_videos_handler(video1, video2)
            # If only one or no videos, return default empty state
            return (
                "Please upload videos or select an example",  # status_output
                gr.Slider(
                    minimum=0, maximum=0, step=1, value=0, interactive=False
                ),  # frame_slider
                None,  # frame1_output
                None,  # frame2_output
                "",  # frame_info (now in metrics_section)
                None,  # ssim_plot
                None,  # psnr_plot
                None,  # mse_plot
                None,  # phash_plot
                None,  # color_plot
                None,  # sharpness_plot
                None,  # overall_plot
                gr.Row(visible=False),  # frame_controls
                gr.Row(visible=False),  # frame_display
                gr.Row(visible=False),  # metrics_section
                gr.Row(visible=False),  # info_section
            )

        # Connect events
        load_btn.click(
            fn=load_videos_handler,
            inputs=[video1_input, video2_input],
            outputs=[
                status_output,
                frame_slider,
                frame1_output,
                frame2_output,
                frame_info,
                ssim_plot,
                psnr_plot,
                mse_plot,
                phash_plot,
                color_plot,
                sharpness_plot,
                overall_plot,
                frame_controls,
                frame_display,
                metrics_section,
                info_section,
            ],
        )

        # Auto-load when both video inputs change (triggered by examples)
        video1_input.change(
            fn=auto_load_when_examples_change,
            inputs=[video1_input, video2_input],
            outputs=[
                status_output,
                frame_slider,
                frame1_output,
                frame2_output,
                frame_info,
                ssim_plot,
                psnr_plot,
                mse_plot,
                phash_plot,
                color_plot,
                sharpness_plot,
                overall_plot,
                frame_controls,
                frame_display,
                metrics_section,
                info_section,
            ],
        )

        video2_input.change(
            fn=auto_load_when_examples_change,
            inputs=[video1_input, video2_input],
            outputs=[
                status_output,
                frame_slider,
                frame1_output,
                frame2_output,
                frame_info,
                ssim_plot,
                psnr_plot,
                mse_plot,
                phash_plot,
                color_plot,
                sharpness_plot,
                overall_plot,
                frame_controls,
                frame_display,
                metrics_section,
                info_section,
            ],
        )

        frame_slider.change(
            fn=update_frames,
            inputs=[frame_slider],
            outputs=[
                frame1_output,
                frame2_output,
                frame_info,
                ssim_plot,
                psnr_plot,
                mse_plot,
                phash_plot,
                color_plot,
                sharpness_plot,
                overall_plot,
            ],
        )

        # Add comprehensive usage guide
        with gr.Accordion("📖 Usage Guide & Metrics Reference", open=False):
            with gr.Accordion("📊 Metrics Explained", open=False):
                gr.Markdown("""
- **SSIM**: Structural Similarity (1.0 = identical structure, 0.0 = completely different)
- **PSNR**: Peak Signal-to-Noise Ratio in dB (higher = better quality, less noise)
- **MSE**: Mean Squared Error (lower = more similar pixel values)
- **pHash**: Perceptual Hash similarity (1.0 = visually identical)
- **Color Histogram**: Color distribution correlation (1.0 = identical color patterns)
- **Sharpness**: Laplacian variance per video (higher = sharper/more detailed images)
- **Overall Quality**: Combined metric averaging SSIM, normalized PSNR, and pHash (when available)
                """)

            with gr.Accordion(
                "🎯 Quality Assessment Scale (Research-Based Thresholds)", open=False
            ):
                gr.Markdown("""
**SSIM Scale** (based on human perception studies):
- 🟢 **Excellent (≥0.9)**: Visually indistinguishable differences
- 🔵 **Good (≥0.8)**: Minor visible differences, still high quality
- 🟡 **Fair (≥0.6)**: Noticeable differences, acceptable quality
- 🔴 **Poor (<0.6)**: Significant visible artifacts and differences

**PSNR Scale** (standard video quality benchmarks):
- 🟢 **Excellent (≥40dB)**: Professional broadcast quality
- 🔵 **Good (≥30dB)**: High consumer video quality
- 🟡 **Fair (≥20dB)**: Acceptable for web streaming
- 🔴 **Poor (<20dB)**: Low quality with visible compression artifacts

**MSE Scale** (pixel difference thresholds):
- 🟢 **Very Similar (≤50)**: Minimal pixel-level differences
- 🔵 **Similar (≤100)**: Small differences, good quality preservation
- 🟡 **Moderately Different (≤200)**: Noticeable but acceptable differences
- 🔴 **Very Different (>200)**: Significant pixel-level changes
                """)

            with gr.Accordion("🔍 Understanding Comparisons", open=False):
                gr.Markdown("""
**Comparison Analysis**: Shows how similar/different the videos are
- Most metrics indicate **similarity** - not which video "wins"
- Higher SSIM/PSNR/pHash/Color = more similar videos
- Lower MSE = more similar videos

**Individual Quality**: Shows the quality of each video separately
- Sharpness comparison shows which video has more detail
- Significance levels: 🔴 MAJOR (>20%), 🟡 MODERATE (10-20%), 🔵 MINOR (5-10%), 🟢 NEGLIGIBLE (<5%)

**Overall Quality**: Combines multiple metrics to provide a single similarity score
- **Formula**: Average of [SSIM + normalized_PSNR + pHash]
- **PSNR Normalization**: PSNR values are divided by 50dB and capped at 1.0
- **Range**: 0.0 to 1.0 (higher = more similar videos overall)
- **Purpose**: Provides a single metric when you need one overall assessment
- **Limitation**: Different metrics may disagree; check individual metrics for details
                """)

            with gr.Accordion("📁 Configuration", open=False):
                gr.Markdown(
                    f"{'Loaded ' + str(len(example_pairs)) + ' example comparisons from data.json' if example_pairs else 'No examples found in data.json'}<br>"
                    f"{'Available videos: ' + str(len(all_videos)) + ' files' if all_videos else ''}"
                )

    return app


def main():
    app = create_app()
    app.launch(server_name="0.0.0.0", server_port=7860, share=False, debug=True)


if __name__ == "__main__":
    main()