Add Gradio visualizer with pipeline data

.gitattributes

@@ -33,3 +33,6 @@ 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
+*.png filter=lfs diff=lfs merge=lfs -text
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.json filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,267 @@
+"""
+Dynamic Intelligence - Human Demo Visualizer
+Egocentric hand tracking dataset visualizer for robot training data
+"""
+
+import gradio as gr
+import json
+import os
+from pathlib import Path
+
+# Load pipeline data
+DATA_DIR = Path(__file__).parent / "data"
+
+def load_json_safe(path):
+    try:
+        with open(path, 'r') as f:
+            return json.load(f)
+    except:
+        return {}
+
+# Load data
+metadata = load_json_safe(DATA_DIR / "metadata.json")
+end_effector = load_json_safe(DATA_DIR / "end_effector.json")
+actions = load_json_safe(DATA_DIR / "actions.json")
+hands_2d = load_json_safe(DATA_DIR / "hands_2d.json")
+
+# Stats
+total_frames = len(metadata.get('poses', []))
+fps = metadata.get('fps', 60)
+hand_detection_rate = len(hands_2d) / max(1, total_frames) * 100
+left_poses = sum(1 for f in end_effector.values() if f.get('left_hand'))
+right_poses = sum(1 for f in end_effector.values() if f.get('right_hand'))
+
+def get_frame_info(frame_idx):
+    """Get info for a specific frame."""
+    frame_key = str(frame_idx)
+    
+    # Hand detection status
+    hand_data = hands_2d.get(frame_key, {})
+    left_detected = hand_data.get('left_hand') is not None
+    right_detected = hand_data.get('right_hand') is not None
+    
+    # End effector pose
+    ee_data = end_effector.get(frame_key, {})
+    left_pose = ee_data.get('left_hand', {}).get('pose_6dof', None)
+    right_pose = ee_data.get('right_hand', {}).get('pose_6dof', None)
+    
+    # Action
+    action_data = actions.get(frame_key, {})
+    left_action = action_data.get('left_hand_action', None)
+    camera_action = action_data.get('camera_action', None)
+    
+    # Format output
+    info = f"""### Frame {frame_idx} / {total_frames - 1}
+**Time:** {frame_idx / fps:.2f}s
+
+---
+
+#### 🖐️ Hand Detection
+- **Left Hand:** {'✅ Detected' if left_detected else '❌ Not Detected'}
+- **Right Hand:** {'✅ Detected' if right_detected else '❌ Not Detected'}
+
+---
+
+#### 📍 End-Effector Pose (6DoF)
+"""
+    
+    if left_pose:
+        info += f"""
+**Left Hand:**
+- Position: X={left_pose[0]*100:.1f}cm, Y={left_pose[1]*100:.1f}cm, Z={left_pose[2]*100:.1f}cm
+- Rotation: Roll={left_pose[3]*57.3:.1f}°, Pitch={left_pose[4]*57.3:.1f}°, Yaw={left_pose[5]*57.3:.1f}°
+"""
+    else:
+        info += "\n**Left Hand:** No pose available\n"
+    
+    if right_pose:
+        info += f"""
+**Right Hand:**
+- Position: X={right_pose[0]*100:.1f}cm, Y={right_pose[1]*100:.1f}cm, Z={right_pose[2]*100:.1f}cm
+- Rotation: Roll={right_pose[3]*57.3:.1f}°, Pitch={right_pose[4]*57.3:.1f}°, Yaw={right_pose[5]*57.3:.1f}°
+"""
+    
+    info += "\n---\n\n#### 🎯 Actions (Delta per frame)\n"
+    
+    if left_action:
+        mag = (left_action[0]**2 + left_action[1]**2 + left_action[2]**2)**0.5 * 100
+        info += f"**Left Hand Movement:** {mag:.2f} cm\n"
+    
+    if camera_action:
+        cam_mag = (camera_action[0]**2 + camera_action[1]**2 + camera_action[2]**2)**0.5 * 100
+        info += f"**Camera Movement:** {cam_mag:.2f} cm\n"
+    
+    return info
+
+def get_frame_image(frame_idx):
+    """Get RGB frame image path."""
+    frame_path = DATA_DIR / "frames" / f"{frame_idx}.jpg"
+    if frame_path.exists():
+        return str(frame_path)
+    return None
+
+def update_display(frame_idx):
+    """Update frame display."""
+    img = get_frame_image(int(frame_idx))
+    info = get_frame_info(int(frame_idx))
+    return img, info
+
+# Build Gradio Interface
+with gr.Blocks(
+    title="DI Human Demo Visualizer",
+    theme=gr.themes.Soft(
+        primary_hue="cyan",
+        secondary_hue="purple",
+    ),
+    css="""
+    .header { text-align: center; margin-bottom: 20px; }
+    .stat-box { 
+        background: linear-gradient(135deg, #1a1a2e, #16213e); 
+        padding: 20px; 
+        border-radius: 12px; 
+        text-align: center;
+        color: white;
+    }
+    .stat-value { font-size: 32px; font-weight: bold; color: #00d4ff; }
+    .stat-label { font-size: 14px; color: #888; }
+    """
+) as demo:
+    
+    # Header
+    gr.Markdown("""
+    # 🤖 Dynamic Intelligence - Human Demo Visualizer
+    
+    **Egocentric hand tracking dataset for humanoid robot training**
+    
+    This visualizer shows the processed output of our LiDAR-based hand tracking pipeline:
+    - RGB video from iPhone LiDAR camera
+    - 21 hand joints tracked via MediaPipe
+    - 6DoF end-effector pose computed from 3 key joints
+    - Actions (delta movements) for robot learning
+    """)
+    
+    # Stats row
+    with gr.Row():
+        with gr.Column(scale=1):
+            gr.Markdown(f"""
+            <div class="stat-box">
+                <div class="stat-value">{total_frames}</div>
+                <div class="stat-label">Total Frames</div>
+            </div>
+            """)
+        with gr.Column(scale=1):
+            gr.Markdown(f"""
+            <div class="stat-box">
+                <div class="stat-value">{hand_detection_rate:.1f}%</div>
+                <div class="stat-label">Hand Detection Rate</div>
+            </div>
+            """)
+        with gr.Column(scale=1):
+            gr.Markdown(f"""
+            <div class="stat-box">
+                <div class="stat-value">{left_poses}</div>
+                <div class="stat-label">Left Hand Poses</div>
+            </div>
+            """)
+        with gr.Column(scale=1):
+            gr.Markdown(f"""
+            <div class="stat-box">
+                <div class="stat-value">{fps}</div>
+                <div class="stat-label">FPS</div>
+            </div>
+            """)
+    
+    gr.Markdown("---")
+    
+    # Main content
+    with gr.Row():
+        # Left: Video frame
+        with gr.Column(scale=2):
+            gr.Markdown("### 📹 RGB Frame")
+            frame_image = gr.Image(
+                label="Frame",
+                type="filepath",
+                height=400
+            )
+            frame_slider = gr.Slider(
+                minimum=0,
+                maximum=total_frames - 1,
+                step=1,
+                value=0,
+                label="Frame",
+                info="Drag to scrub through video"
+            )
+        
+        # Right: Frame info
+        with gr.Column(scale=1):
+            frame_info = gr.Markdown(get_frame_info(0))
+    
+    gr.Markdown("---")
+    
+    # Plots
+    gr.Markdown("### 📊 Validation Plots")
+    
+    with gr.Row():
+        gr.Image(
+            value=str(DATA_DIR / "plots" / "camera_trajectory.png"),
+            label="Camera Trajectory (World Frame)",
+            height=300
+        )
+        gr.Image(
+            value=str(DATA_DIR / "plots" / "left_hand_trajectory.png"),
+            label="Left Hand Trajectory (Camera Frame)",
+            height=300
+        )
+    
+    with gr.Row():
+        gr.Image(
+            value=str(DATA_DIR / "plots" / "hand_pose_vs_time.png"),
+            label="Hand 6DoF Pose vs Time",
+            height=300
+        )
+        gr.Image(
+            value=str(DATA_DIR / "plots" / "actions_histogram.png"),
+            label="Actions Distribution",
+            height=300
+        )
+    
+    # Physics validation
+    gr.Markdown("""
+    ---
+    
+    ### ✅ Physics Validation Results
+    
+    | Check | Status | Details |
+    |-------|--------|---------|
+    | Camera Trajectory | ✅ PASS | Smooth movement, ~40cm total range |
+    | Hand Depth Range | ✅ PASS | 15-60cm from camera (realistic) |
+    | Action Magnitudes | ✅ PASS | Median 0.34cm/frame (no tracking errors) |
+    | Outlier Detection | ✅ PASS | No large jumps detected |
+    | 6DoF Rotations | ✅ PASS | Natural hand movement patterns |
+    
+    ---
+    
+    ### 📋 Dataset Info
+    
+    - **Source:** iPhone 12 Pro+ LiDAR camera (Record3D app)
+    - **Pipeline:** MediaPipe hand tracking → Depth lookup → 6DoF computation
+    - **Format:** Compatible with Physical Intelligence / LeRobot training
+    - **Organization:** [Dynamic Intelligence](https://huggingface.co/DynamicIntelligence)
+    """)
+    
+    # Event handler
+    frame_slider.change(
+        fn=update_display,
+        inputs=[frame_slider],
+        outputs=[frame_image, frame_info]
+    )
+    
+    # Initialize
+    demo.load(
+        fn=lambda: update_display(0),
+        outputs=[frame_image, frame_info]
+    )
+
+if __name__ == "__main__":
+    demo.launch()
+

data/actions.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69aa1206d44f13f794e5cb9f9099a1f32b4e6d95490a7e0be58a9826ce3f8b1b
+size 534384

data/end_effector.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efdca95bf448637843f7b1bc756de9ecaa76a1d8f92fe5e284112d34a60995f9
+size 505857