import os import numpy as np import plotly.graph_objects as go from action_state.utils import ( CO3DDataLoader, get_camera_center, get_view_direction, get_camera_up, get_sequence_geometry ) # --- 配置 --- ROOT_PATH = "/run/determined/NAS1/public/lixinyuan/interleaved-co3d" CATEGORY = "tv" SEQUENCE_NAME = "398_50483_99041" OUTPUT_FILE = "interactive_3d_aligned.html" HIGHLIGHT_FRAMES = [74, 85, 157, 98, 116] def main(): print(f"Loading sequence: {SEQUENCE_NAME}...") loader = CO3DDataLoader(ROOT_PATH, CATEGORY) if SEQUENCE_NAME not in loader.get_sequences(): print(f"Error: Sequence {SEQUENCE_NAME} not found.") return frame_ids = sorted(loader.get_frames(SEQUENCE_NAME)) seq_data = loader.seq_data[SEQUENCE_NAME] # 使用CO3D官方对齐方法 mean_center, basis, aligned_seq_data = get_sequence_geometry( seq_data, align_to_standard=True ) print(f"\n{'='*60}") print(f"Aligned to Standard Y-Up Coordinate System") print(f"Original CO3D ground normal → Standard [0,1,0]") print(f"{'='*60}\n") # 收集对齐后的相机数据 camera_centers = [] view_dirs = [] up_vectors = [] for fid in frame_ids: info = aligned_seq_data[fid] C = get_camera_center(info['R'], info['T']) V = get_view_direction(info['R']) U = get_camera_up(info['R']) camera_centers.append(C) view_dirs.append(V) up_vectors.append(U) camera_centers = np.array(camera_centers) view_dirs = np.array(view_dirs) up_vectors = np.array(up_vectors) # 计算统计 dists = np.linalg.norm(camera_centers - mean_center, axis=1) avg_dist = np.mean(dists) arrow_len = avg_dist * 0.3 # 分析Up向量对齐度 mean_camera_up = np.mean(up_vectors, axis=0) mean_camera_up_normalized = mean_camera_up / np.linalg.norm(mean_camera_up) alignment = np.dot(mean_camera_up_normalized, basis[2]) # === 创建Plotly图形 === fig = go.Figure() # 1. 地面平面（XZ平面） plane_size = avg_dist * 2 x_range = np.linspace(mean_center[0] - plane_size, mean_center[0] + plane_size, 20) z_range = np.linspace(mean_center[2] - plane_size, mean_center[2] + plane_size, 20) X_grid, Z_grid = np.meshgrid(x_range, z_range) Y_grid = np.ones_like(X_grid) * mean_center[1] fig.add_trace(go.Surface( x=X_grid, y=Y_grid, z=Z_grid, colorscale=[[0, 'lightblue'], [1, 'lightblue']], showscale=False, opacity=0.15, name='Ground Plane (XZ)', hoverinfo='skip' )) # 2. 世界坐标系轴 axis_len = avg_dist * 0.5 # X轴 - 红色 fig.add_trace(go.Scatter3d( x=[mean_center[0], mean_center[0] + axis_len], y=[mean_center[1], mean_center[1]], z=[mean_center[2], mean_center[2]], mode='lines+markers', line=dict(color='red', width=8), marker=dict(size=[0, 10], symbol='arrow', angleref='previous'), name='World X', hovertemplate='X Axis' )) # Y轴 - 绿色（Up，对齐后的地面法向量） fig.add_trace(go.Scatter3d( x=[mean_center[0], mean_center[0]], y=[mean_center[1], mean_center[1] + axis_len], z=[mean_center[2], mean_center[2]], mode='lines+markers', line=dict(color='lime', width=8), marker=dict(size=[0, 10], symbol='arrow', angleref='previous'), name='World Y (Up/Ground Normal)', hovertemplate='Y Axis (Up)
Aligned from CO3D ground normal' )) # Z轴 - 蓝色 fig.add_trace(go.Scatter3d( x=[mean_center[0], mean_center[0]], y=[mean_center[1], mean_center[1]], z=[mean_center[2], mean_center[2] + axis_len], mode='lines+markers', line=dict(color='cyan', width=8), marker=dict(size=[0, 10], symbol='arrow', angleref='previous'), name='World Z', hovertemplate='Z Axis' )) # 3. 物体中心 fig.add_trace(go.Scatter3d( x=[mean_center[0]], y=[mean_center[1]], z=[mean_center[2]], mode='markers', marker=dict(size=15, color='black', symbol='x', line=dict(width=3, color='yellow')), name='Object Center', hovertemplate='Object Center
X: %{x:.2f}
Y: %{y:.2f}
Z: %{z:.2f}' )) # 4. 相机轨迹线 fig.add_trace(go.Scatter3d( x=camera_centers[:, 0], y=camera_centers[:, 1], z=camera_centers[:, 2], mode='lines', line=dict(color='gray', width=2, dash='dash'), name='Camera Trajectory', opacity=0.5, hoverinfo='skip' )) # 5. 所有相机位置 fig.add_trace(go.Scatter3d( x=camera_centers[:, 0], y=camera_centers[:, 1], z=camera_centers[:, 2], mode='markers', marker=dict(size=3, color=frame_ids, colorscale='Viridis', colorbar=dict(title="Frame ID"), opacity=0.6), text=[f'Frame {fid}' for fid in frame_ids], name='All Cameras', hovertemplate='%{text}
X: %{x:.2f}
Y: %{y:.2f}
Z: %{z:.2f}' )) # 6. 高亮帧 highlight_indices = [i for i, fid in enumerate(frame_ids) if fid in HIGHLIGHT_FRAMES] if highlight_indices: highlight_centers = camera_centers[highlight_indices] highlight_fids = [frame_ids[i] for i in highlight_indices] fig.add_trace(go.Scatter3d( x=highlight_centers[:, 0], y=highlight_centers[:, 1], z=highlight_centers[:, 2], mode='markers+text', marker=dict(size=10, color='red', symbol='circle', line=dict(width=2, color='black')), text=[str(fid) for fid in highlight_fids], textposition='top center', textfont=dict(size=12, color='red', family='Arial Black'), name='Highlighted Frames', hovertemplate='Frame %{text}
X: %{x:.2f}
Y: %{y:.2f}
Z: %{z:.2f}' )) # 7. 高亮帧的相机朝向和Up向量 for idx in highlight_indices: C = camera_centers[idx] V = view_dirs[idx] U = up_vectors[idx] # View direction (深蓝色) view_end = C + V * arrow_len fig.add_trace(go.Scatter3d( x=[C[0], view_end[0]], y=[C[1], view_end[1]], z=[C[2], view_end[2]], mode='lines', line=dict(color='darkblue', width=4), name=f'View Dir (Frame {frame_ids[idx]})', showlegend=False, hovertemplate=f'View Direction
Frame {frame_ids[idx]}' )) # Up vector (洋红色) up_end = C + U * arrow_len fig.add_trace(go.Scatter3d( x=[C[0], up_end[0]], y=[C[1], up_end[1]], z=[C[2], up_end[2]], mode='lines', line=dict(color='magenta', width=4), name=f'Up Vec (Frame {frame_ids[idx]})', showlegend=False, hovertemplate=f'Camera Up Vector
Frame {frame_ids[idx]}' )) # 连线到物体中心 fig.add_trace(go.Scatter3d( x=[C[0], mean_center[0]], y=[C[1], mean_center[1]], z=[C[2], mean_center[2]], mode='lines', line=dict(color='black', width=1, dash='dot'), opacity=0.3, showlegend=False, hoverinfo='skip' )) # 8. 平均相机Up向量（用金色虚线表示） mean_up_end = mean_center + mean_camera_up_normalized * axis_len * 0.9 fig.add_trace(go.Scatter3d( x=[mean_center[0], mean_up_end[0]], y=[mean_center[1], mean_up_end[1]], z=[mean_center[2], mean_up_end[2]], mode='lines', line=dict(color='gold', width=6, dash='dash'), name=f'Mean Camera Up (align={alignment:.2f})', hovertemplate=f'Mean Camera Up
Alignment: {alignment:.3f}' )) # 设置布局 fig.update_layout( title=dict( text=f'Interactive 3D Camera Trajectory (CO3D Aligned)
' + f'Sequence: {CATEGORY}/{SEQUENCE_NAME}
' + f'Original CO3D Ground Normal: [-0.0396,-0.8306,-0.5554] → Aligned [0,1,0]
' + f' 0.7 else "red"}">Camera Up Alignment: {alignment:.3f} {"✓" if abs(alignment) > 0.7 else "✗"}', x=0.5, xanchor='center' ), scene=dict( xaxis=dict( title='X (PyTorch3D: Left)', backgroundcolor="rgb(240, 240, 240)", gridcolor="white", showbackground=True ), yaxis=dict( title='Y (PyTorch3D: Up)', backgroundcolor="rgb(230, 255, 230)", gridcolor="white", showbackground=True ), zaxis=dict( title='Z (PyTorch3D: Forward)', backgroundcolor="rgb(240, 240, 240)", gridcolor="white", showbackground=True ), aspectmode='data', camera=dict( eye=dict(x=1.5, y=1.5, z=1.5), center=dict(x=0, y=0, z=0), up=dict(x=0, y=1, z=0) ) ), hovermode='closest', width=1400, height=900, showlegend=True, legend=dict( x=0.02, y=0.98, bgcolor='rgba(255,255,255,0.9)', bordercolor='black', borderwidth=1 ) ) # 添加注释说明 fig.add_annotation( text="🖱️ Controls:
" + "• Drag to rotate
" + "• Scroll to zoom
" + "• Hover for details
" + "• Double-click to reset", xref="paper", yref="paper", x=0.98, y=0.02, xanchor='right', yanchor='bottom', showarrow=False, bgcolor='rgba(255,255,200,0.8)', bordercolor='black', borderwidth=1, font=dict(size=10) ) # 保存为HTML fig.write_html(OUTPUT_FILE) print(f"\n✅ Interactive 3D plot saved to {OUTPUT_FILE}") print(f"📂 Open it in your browser to interact!") print(f"\n📊 Analysis:") print(f" Camera Up Alignment: {alignment:.3f}") print(f" Status: {'✓ Well aligned' if abs(alignment) > 0.7 else '✗ Misaligned'}") print(f" Total Frames: {len(frame_ids)}") print(f" Highlighted Frames: {len(highlight_indices)}") if __name__ == "__main__": main()