Push Space to HF

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+pose_landmarker.task filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,19 @@
+FROM python:3.10-slim
+
+WORKDIR /app
+
+RUN apt-get update && apt-get install -y \
+    ffmpeg \
+    libgl1 \
+    libglib2.0-0 \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt .
+
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+ENV PORT=7860
+
+CMD ["python", "app.py"]

README.md

@@ -1,11 +1,58 @@
 ---
 title: MediaPipe Pose Estimation
-emoji: 🐢
-colorFrom: gray
-colorTo: red
+emoji: 🧍
+colorFrom: blue
+colorTo: purple
 sdk: docker
 pinned: false
-short_description: Track 33 body keypoints using MediaPipe
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# MediaPipe Pose Estimation
+
+Professional pose detection and visualization system powered by Google MediaPipe. Track 33 body keypoints with customizable visualization options.
+
+## Features
+
+- Real-time pose estimation with 33 anatomical keypoints
+- Adjustable detection and tracking confidence thresholds
+- Black background or original video overlay options
+- Single color or multicolor body part visualization
+- Professional-grade output with H.264 encoding
+
+## Quick Start
+
+1. Upload a video file (MP4, AVI, MOV, WebM)
+2. Adjust confidence parameters (default: 0.5 recommended)
+3. Select background type and color mode
+4. Process video
+
+## Parameters
+
+**Detection Confidence** (0.0-1.0)  
+Controls pose detection sensitivity. Higher values reduce false positives.
+
+**Tracking Confidence** (0.0-1.0)  
+Determines tracking stability across frames. Higher values provide more stable tracking.
+
+## Color Modes
+
+**Single Color**: Uniform skeleton visualization with customizable colors
+
+**Multicolor**: Body parts colored by region
+- Torso: Yellow
+- Right Arm: Red | Left Arm: Blue  
+- Right Leg: Magenta | Left Leg: Green
+
+## Output
+
+- Format: MP4 (H.264)
+- Resolution and frame rate match input
+- Processing time scales with video length and resolution
+
+## Technical Stack
+
+MediaPipe Pose Landmarker | OpenCV | Gradio | Python
+
+---
+
+**Note**: Videos exceeding 2 minutes or high resolution may require extended processing time.

app.py

@@ -0,0 +1,484 @@
+import cv2
+import numpy as np
+import gradio as gr
+import mediapipe as mp
+from mediapipe.tasks.python import vision
+from mediapipe.tasks.python import BaseOptions
+from mediapipe.tasks.python.vision import PoseLandmarker, PoseLandmarkerOptions, RunningMode
+
+MODEL_PATH = "pose_landmarker.task"
+
+# Color palette (BGR format)
+COLORS = {
+    "White": (255, 255, 255),
+    "Red": (0, 0, 255),
+    "Green": (0, 255, 0),
+    "Blue": (255, 0, 0),
+    "Yellow": (0, 255, 255),
+    "Cyan": (255, 255, 0),
+    "Magenta": (255, 0, 255),
+    "Orange": (0, 165, 255),
+    "Purple": (255, 0, 128),
+    "Pink": (203, 192, 255),
+}
+
+MULTICOLOR_SCHEME = {
+    "face": (255, 255, 0),        # Cyan
+    "torso": (0, 255, 255),       # Yellow
+    "right_arm": (0, 0, 255),     # Red
+    "left_arm": (255, 0, 0),      # Blue
+    "right_leg": (255, 0, 255),   # Magenta
+    "left_leg": (0, 255, 0),      # Green
+}
+
+def get_body_part_connections():
+    """Define which connections belong to which body part"""
+    
+    connections = {
+        "face": [
+            (0, 1), (1, 2), (2, 3), (3, 7),  # Right eye region
+            (0, 4), (4, 5), (5, 6), (6, 8),  # Left eye region
+            (9, 10),  # Mouth
+        ],
+        "torso": [
+            (11, 12),  # Shoulders
+            (11, 23), (12, 24),  # Shoulder to hip
+            (23, 24),  # Hips
+        ],
+        "right_arm": [
+            (11, 13), (13, 15),  # Shoulder to elbow to wrist
+            (15, 17), (15, 19), (15, 21),  # Wrist connections
+            (17, 19),  # Hand
+        ],
+        "left_arm": [
+            (12, 14), (14, 16),  # Shoulder to elbow to wrist
+            (16, 18), (16, 20), (16, 22),  # Wrist connections
+            (18, 20),  # Hand
+        ],
+        "right_leg": [
+            (23, 25), (25, 27),  # Hip to knee to ankle
+            (27, 29), (27, 31),  # Ankle connections
+            (29, 31),  # Foot
+        ],
+        "left_leg": [
+            (24, 26), (26, 28),  # Hip to knee to ankle
+            (28, 30), (28, 32),  # Ankle connections
+            (30, 32),  # Foot
+        ],
+    }
+    return connections
+
+def draw_pose(
+    video_path,
+    detection_confidence,
+    tracking_confidence,
+    background_type,
+    color_mode,
+    line_color,
+    joint_color
+):
+    output_path = "output.mp4"
+    
+    options = PoseLandmarkerOptions(
+        base_options=BaseOptions(model_asset_path=MODEL_PATH),
+        running_mode=RunningMode.VIDEO,
+        num_poses=1,
+        min_pose_detection_confidence=detection_confidence,
+        min_tracking_confidence=tracking_confidence,
+    )
+    landmarker = PoseLandmarker.create_from_options(options)
+    
+    cap = cv2.VideoCapture(video_path)
+    width = int(cap.get(3))
+    height = int(cap.get(4))
+    fps = cap.get(5) or 24
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    out = cv2.VideoWriter(output_path, fourcc, fps, (width, height))
+    
+    timestamp = 0
+    
+    line_bgr = COLORS[line_color]
+    joint_bgr = COLORS[joint_color]
+    
+
+    body_parts = get_body_part_connections()
+    
+    while cap.isOpened():
+        ret, frame = cap.read()
+        if not ret:
+            break
+        
+        rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        mp_image = mp.Image(
+            image_format=mp.ImageFormat.SRGB,
+            data=rgb_frame
+        )
+        
+        result = landmarker.detect_for_video(mp_image, timestamp)
+        timestamp += int(1000 / fps)
+        
+        if background_type == "Black Background":
+            canvas = np.zeros((height, width, 3), dtype=np.uint8)
+        else:  # Original Video
+            canvas = frame.copy()
+        
+        if result.pose_landmarks:
+            for pose_landmarks in result.pose_landmarks:
+                points = []
+                for lm in pose_landmarks:
+                    x = int(lm.x * width)
+                    y = int(lm.y * height)
+                    points.append((x, y))
+                
+                if color_mode == "Single Color":
+                    # Original behavior - single color for all connections
+                    connections = mp.solutions.pose.POSE_CONNECTIONS
+                    for c in connections:
+                        cv2.line(
+                            canvas,
+                            points[c[0]],
+                            points[c[1]],
+                            line_bgr,
+                            3
+                        )
+                else:  # Multicolor
+                    for part_name, part_connections in body_parts.items():
+                        part_color = MULTICOLOR_SCHEME[part_name]
+                        for c in part_connections:
+                            if c[0] < len(points) and c[1] < len(points):
+                                cv2.line(
+                                    canvas,
+                                    points[c[0]],
+                                    points[c[1]],
+                                    part_color,
+                                    3
+                                )
+                
+                for p in points:
+                    cv2.circle(canvas, p, 5, joint_bgr, -1)
+        
+        out.write(canvas)
+    
+    cap.release()
+    out.release()
+    
+    return output_path
+
+custom_css = """
+* {
+    font-family: 'Inter', 'Segoe UI', system-ui, -apple-system, sans-serif;
+}
+
+.main-header {
+    text-align: center;
+    padding: 30px 20px;
+    border-bottom: 1px solid #404040;
+    margin-bottom: 30px;
+}
+
+.main-header h1 {
+    font-size: 28px;
+    font-weight: 600;
+    color: #ffffff;
+    margin: 0 0 8px 0;
+    letter-spacing: -0.5px;
+}
+
+.main-header p {
+    font-size: 15px;
+    color: #b0b0b0;
+    margin: 0;
+    font-weight: 400;
+}
+
+.section-header {
+    font-size: 13px;
+    font-weight: 600;
+    color: #ffffff;
+    text-transform: uppercase;
+    letter-spacing: 0.5px;
+    margin-bottom: 16px;
+    padding-bottom: 8px;
+    border-bottom: 2px solid #ffffff;
+}
+
+.subsection-title {
+    font-size: 12px;
+    font-weight: 600;
+    color: #b0b0b0;
+    text-transform: uppercase;
+    letter-spacing: 0.3px;
+    margin: 20px 0 12px 0;
+}
+
+.warning-box {
+    background: #1a1a1a;
+    border-left: 3px solid #888;
+    padding: 16px 20px;
+    margin: 20px 0;
+    font-size: 14px;
+    color: #d0d0d0;
+    line-height: 1.6;
+}
+
+.warning-box strong {
+    font-weight: 600;
+    color: #ffffff;
+}
+
+.info-box {
+    background: #1a1a1a;
+    border: 1px solid #404040;
+    padding: 16px;
+    border-radius: 4px;
+    margin-top: 16px;
+}
+
+.info-box p {
+    margin: 6px 0;
+    font-size: 13px;
+    color: #b0b0b0;
+    line-height: 1.5;
+}
+
+.info-box strong {
+    color: #ffffff;
+}
+
+.footer {
+    text-align: center;
+    padding: 30px 20px;
+    border-top: 1px solid #404040;
+    margin-top: 40px;
+    color: #888;
+    font-size: 13px;
+}
+
+.footer h3 {
+    font-size: 14px;
+    font-weight: 600;
+    color: #ffffff;
+    margin-bottom: 12px;
+}
+
+.footer p {
+    color: #b0b0b0;
+}
+
+#submit-btn {
+    margin-top: 24px;
+    background: #ffffff;
+    color: #000000;
+    border: none;
+    font-weight: 500;
+    letter-spacing: 0.3px;
+}
+
+#submit-btn:hover {
+    background: #e0e0e0;
+}
+
+.gr-box {
+    border-radius: 4px;
+}
+
+.gr-input, .gr-dropdown, .gr-radio {
+    border-radius: 4px;
+}
+
+.gr-accordion {
+    border: 1px solid #404040;
+    border-radius: 4px;
+}
+
+label {
+    color: #d0d0d0 !important;
+}
+
+.gr-text-input, .gr-dropdown {
+    background: #1a1a1a;
+    border: 1px solid #404040;
+    color: #ffffff;
+}
+
+.color-legend {
+    background: #1a1a1a;
+    border: 1px solid #404040;
+    padding: 12px;
+    border-radius: 4px;
+    margin-top: 12px;
+    font-size: 12px;
+}
+
+.color-legend p {
+    margin: 4px 0;
+    color: #b0b0b0;
+}
+
+.color-item {
+    display: inline-block;
+    width: 12px;
+    height: 12px;
+    margin-right: 6px;
+    border-radius: 2px;
+}
+"""
+
+with gr.Blocks(title="MediaPipe Pose Estimation", theme=gr.themes.Default(), css=custom_css) as demo:
+    
+    # Header
+    gr.HTML(
+        """
+        <div class="main-header">
+            <h1>MediaPipe Pose Estimation</h1>
+            <p>Advanced pose detection and visualization system</p>
+        </div>
+        """
+    )
+    
+    with gr.Accordion("Instructions", open=False):
+        gr.Markdown(
+            """
+            1. Upload a video file containing human subjects
+            2. Adjust detection and tracking confidence parameters as needed
+            3. Select the desired background type for the output
+            4. Choose between single color or multicolor visualization
+            5. If single color is selected, customize the skeleton colors
+            6. Click Process Video to generate the result
+            """
+        )
+    
+    with gr.Accordion("Parameter Guidelines", open=False):
+        gr.Markdown(
+            """
+            **Detection Confidence**
+            
+            Controls the minimum confidence threshold for initial pose detection. Higher values (0.7-0.9) reduce false positives but may miss some poses. Lower values (0.3-0.5) detect more poses but may include incorrect detections. Default: 0.5
+            
+            **Tracking Confidence**
+            
+            Determines the reliability threshold for tracking poses across consecutive frames. Higher values provide more stable tracking but may lose poses more easily. Lower values maintain tracking longer but may be less stable. Default: 0.5
+            
+            **Color Modes**
+            
+            Single Color: All skeleton lines use the same color. Multicolor: Different body parts are colored differently (face, torso, arms, legs).
+            """
+        )
+    
+    # Warning message
+    gr.HTML(
+        """
+        <div class="warning-box">
+            <strong>Processing Time Notice:</strong> Processing duration is proportional to video length and resolution. Videos exceeding 2 minutes or high-resolution files may require several minutes to process. Please wait while the system completes the analysis.
+        </div>
+        """
+    )
+    
+    with gr.Row():
+        with gr.Column():
+            gr.HTML('<div class="section-header">Input</div>')
+            
+            video_input = gr.Video(label="Video File")
+            
+            gr.HTML('<div class="subsection-title">Confidence Parameters</div>')
+            detection_conf = gr.Slider(
+                minimum=0.0,
+                maximum=1.0,
+                value=0.5,
+                step=0.05,
+                label="Detection Confidence"
+            )
+            tracking_conf = gr.Slider(
+                minimum=0.0,
+                maximum=1.0,
+                value=0.5,
+                step=0.05,
+                label="Tracking Confidence"
+            )
+            
+            gr.HTML('<div class="subsection-title">Visualization Options</div>')
+            background_type = gr.Radio(
+                choices=["Black Background", "Original Video"],
+                value="Black Background",
+                label="Background Type"
+            )
+            
+            color_mode = gr.Radio(
+                choices=["Single Color", "Multicolor"],
+                value="Single Color",
+                label="Color Mode"
+            )
+            
+            with gr.Row():
+                line_color = gr.Dropdown(
+                    choices=list(COLORS.keys()),
+                    value="White",
+                    label="Line Color",
+                    info="Only applies in Single Color mode"
+                )
+                joint_color = gr.Dropdown(
+                    choices=list(COLORS.keys()),
+                    value="Red",
+                    label="Joint Color",
+                    info='Single & Multicolor'
+                )
+            
+            gr.HTML(
+                """
+                <div class="color-legend">
+                    <p><strong>Multicolor Legend:</strong></p>
+                    <p><span class="color-item" style="background: rgb(255, 255, 0);"></span>Torso: Yellow</p>
+                    <p><span class="color-item" style="background: rgb(255, 0, 0);"></span>Right Arm: Red</p>
+                    <p><span class="color-item" style="background: rgb(0, 0, 255);"></span>Left Arm: Blue</p>
+                    <p><span class="color-item" style="background: rgb(255, 0, 255);"></span>Right Leg: Magenta</p>
+                    <p><span class="color-item" style="background: rgb(0, 255, 0);"></span>Left Leg: Green</p>
+                </div>
+                """
+            )
+            
+            submit_btn = gr.Button("Process Video", variant="primary", elem_id="submit-btn")
+        
+        with gr.Column():
+            gr.HTML('<div class="section-header">Output</div>')
+            video_output = gr.Video(label="Processed Video")
+            
+            gr.HTML(
+                """
+                <div class="info-box">
+                    <p><strong>Output Specifications:</strong></p>
+                    <p>Format: MP4 (H.264 encoding)</p>
+                    <p>Resolution: Matches input resolution</p>
+                    <p>Frame Rate: Matches input frame rate</p>
+                    <p>Keypoints: 33 body landmarks tracked per frame</p>
+                </div>
+                """
+            )
+    
+    # Footer
+    gr.HTML(
+        """
+        <div class="footer">
+            <h3>Technical Information</h3>
+            <p>This application utilizes MediaPipe Pose Landmarker for real-time pose detection and tracking.</p>
+            <p>The system identifies 33 anatomical keypoints and visualizes skeletal structure with customizable styling.</p>
+            <p style="margin-top: 16px;">Supported formats: MP4, AVI, MOV, WebM</p>
+            <p style="margin-top: 16px; color: #666;">Powered by Google MediaPipe</p>
+        </div>
+        """
+    )
+    
+    submit_btn.click(
+        fn=draw_pose,
+        inputs=[
+            video_input,
+            detection_conf,
+            tracking_conf,
+            background_type,
+            color_mode,
+            line_color,
+            joint_color
+        ],
+        outputs=video_output
+    )
+
+demo.launch(server_name="0.0.0.0", server_port=7860)

pose_landmarker.task

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5134a3aad27a58b93da0088d431f366da362b44e3ccfbe3462b3827a839011b1
+size 9398198

requirements.txt

@@ -0,0 +1,4 @@
+mediapipe==0.10.21
+opencv-python-headless==4.11.0.86
+gradio
+numpy==1.26.4