Upload scripts/visualize_canonical.py with huggingface_hub

scripts/visualize_canonical.py

@@ -0,0 +1,297 @@
+"""
+Visualize canonical naming, side tags, and symmetry pairs across all datasets.
+
+Produces:
+1. Per-dataset skeleton with joints colored by side tag (L=red, R=blue, C=gray)
+2. Canonical name labels on joints
+3. Symmetry pairs connected by dashed lines
+4. Cross-dataset canonical consistency heatmap
+"""
+
+import sys
+import os
+from pathlib import Path
+import numpy as np
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+import matplotlib.patches as mpatches
+
+project_root = Path(__file__).parent.parent
+sys.path.insert(0, str(project_root))
+
+from src.data.skeleton_graph import SkeletonGraph
+from scripts.preprocess_bvh import forward_kinematics
+
+RESULT_DIR = project_root / 'results' / 'canonical_check'
+RESULT_DIR.mkdir(parents=True, exist_ok=True)
+
+SIDE_COLORS = {'left': '#e74c3c', 'right': '#3498db', 'center': '#95a5a6'}
+
+
+def plot_skeleton_canonical(ax, positions, parents, side_tags, canonical_names,
+                            symmetry_pairs, title, label_fontsize=6):
+    """Plot skeleton with side-colored joints and canonical name labels."""
+    J = len(parents)
+    pos = positions
+
+    # Draw bones
+    for j in range(J):
+        p = parents[j]
+        if p >= 0:
+            ax.plot3D([pos[j, 0], pos[p, 0]], [pos[j, 2], pos[p, 2]],
+                     [pos[j, 1], pos[p, 1]], color='#bdc3c7', linewidth=1.5, zorder=1)
+
+    # Draw symmetry pairs as dashed lines
+    for i, j in symmetry_pairs:
+        if i < J and j < J:
+            mid_y = (pos[i, 1] + pos[j, 1]) / 2
+            ax.plot3D([pos[i, 0], pos[j, 0]], [pos[i, 2], pos[j, 2]],
+                     [pos[i, 1], pos[j, 1]], color='#2ecc71', linewidth=0.8,
+                     linestyle='--', alpha=0.5, zorder=2)
+
+    # Draw joints colored by side
+    for j in range(J):
+        color = SIDE_COLORS.get(side_tags[j], '#95a5a6')
+        ax.scatter3D([pos[j, 0]], [pos[j, 2]], [pos[j, 1]],
+                    color=color, s=30, zorder=3, edgecolors='black', linewidths=0.3)
+
+    # Label joints with canonical names (subsample if too many)
+    step = max(1, J // 15)
+    for j in range(0, J, step):
+        ax.text(pos[j, 0], pos[j, 2], pos[j, 1] + 0.02,
+               canonical_names[j], fontsize=label_fontsize, ha='center', va='bottom',
+               color='black', alpha=0.8)
+
+    ax.set_title(title, fontsize=10, fontweight='bold')
+    ax.set_xlabel('X', fontsize=7)
+    ax.set_ylabel('Z', fontsize=7)
+    ax.set_zlabel('Y', fontsize=7)
+    ax.tick_params(labelsize=5)
+
+    # Equal axes
+    mid = pos.mean(axis=0)
+    span = max(pos.max(axis=0) - pos.min(axis=0)) / 2 + 0.05
+    ax.set_xlim(mid[0] - span, mid[0] + span)
+    ax.set_ylim(mid[2] - span, mid[2] + span)
+    ax.set_zlim(mid[1] - span, mid[1] + span)
+
+
+def get_rest_pose(dataset_path, dataset_id):
+    """Get rest pose joint positions from first motion."""
+    motions_dir = dataset_path / 'motions'
+    files = sorted(os.listdir(motions_dir))
+    if not files:
+        return None
+    d = dict(np.load(motions_dir / files[0], allow_pickle=True))
+    return d['joint_positions'][0]  # first frame
+
+
+def visualize_human_datasets():
+    """Plot 1: All 6 human datasets side by side."""
+    fig = plt.figure(figsize=(24, 8))
+    datasets = ['humanml3d', 'lafan1', '100style', 'bandai_namco', 'cmu_mocap', 'mixamo']
+
+    for idx, ds in enumerate(datasets):
+        ds_path = project_root / 'data' / 'processed' / ds
+        s = dict(np.load(ds_path / 'skeleton.npz', allow_pickle=True))
+        sg = SkeletonGraph.from_dict(s)
+        canon = [str(n) for n in s['canonical_names']]
+        rest_pos = get_rest_pose(ds_path, ds)
+        if rest_pos is None:
+            continue
+
+        ax = fig.add_subplot(1, 6, idx + 1, projection='3d')
+        plot_skeleton_canonical(
+            ax, rest_pos, sg.parent_indices, sg.side_tags, canon,
+            sg.symmetry_pairs,
+            f'{ds}\n{sg.num_joints}j, {len(sg.symmetry_pairs)} sym pairs',
+            label_fontsize=5 if sg.num_joints > 30 else 6,
+        )
+
+    # Legend
+    patches = [
+        mpatches.Patch(color=SIDE_COLORS['left'], label='Left'),
+        mpatches.Patch(color=SIDE_COLORS['right'], label='Right'),
+        mpatches.Patch(color=SIDE_COLORS['center'], label='Center'),
+        mpatches.Patch(color='#2ecc71', label='Symmetry pair'),
+    ]
+    fig.legend(handles=patches, loc='lower center', ncol=4, fontsize=9)
+
+    plt.suptitle('Human Datasets — Canonical Names + Side Tags + Symmetry',
+                fontsize=14, fontweight='bold')
+    plt.tight_layout(rect=[0, 0.05, 1, 0.95])
+    out = RESULT_DIR / 'human_canonical_overview.png'
+    plt.savefig(out, dpi=150, bbox_inches='tight')
+    plt.close()
+    print(f'Saved: {out}')
+
+
+def visualize_zoo_animals():
+    """Plot 2: Diverse Zoo animals with canonical names."""
+    species = ['Dog', 'Cat', 'Horse', 'Eagle', 'Trex', 'Spider',
+               'Ant', 'Anaconda', 'Dragon', 'Crab', 'Elephant', 'Bat']
+    fig = plt.figure(figsize=(24, 18))
+
+    zoo_path = project_root / 'data' / 'processed' / 'truebones_zoo'
+    motions_dir = zoo_path / 'motions'
+
+    # Find one motion per species
+    species_motions = {}
+    for f in sorted(os.listdir(motions_dir)):
+        d = dict(np.load(motions_dir / f, allow_pickle=True))
+        sp = str(d.get('species', ''))
+        if sp in species and sp not in species_motions:
+            species_motions[sp] = d
+
+    for idx, sp in enumerate(species):
+        if sp not in species_motions:
+            continue
+        d = species_motions[sp]
+        skel_path = zoo_path / 'skeletons' / f'{sp}.npz'
+        if not skel_path.exists():
+            continue
+
+        s = dict(np.load(skel_path, allow_pickle=True))
+        sg = SkeletonGraph.from_dict(s)
+        canon = [str(n) for n in s['canonical_names']]
+        rest_pos = d['joint_positions'][0]
+
+        ax = fig.add_subplot(3, 4, idx + 1, projection='3d')
+        plot_skeleton_canonical(
+            ax, rest_pos, sg.parent_indices, sg.side_tags, canon,
+            sg.symmetry_pairs,
+            f'{sp} ({sg.num_joints}j, {len(sg.symmetry_pairs)} sym)',
+            label_fontsize=5,
+        )
+
+    patches = [
+        mpatches.Patch(color=SIDE_COLORS['left'], label='Left'),
+        mpatches.Patch(color=SIDE_COLORS['right'], label='Right'),
+        mpatches.Patch(color=SIDE_COLORS['center'], label='Center'),
+        mpatches.Patch(color='#2ecc71', label='Symmetry pair'),
+    ]
+    fig.legend(handles=patches, loc='lower center', ncol=4, fontsize=10)
+
+    plt.suptitle('Truebones Zoo — Canonical Names + Side Tags + Symmetry',
+                fontsize=14, fontweight='bold')
+    plt.tight_layout(rect=[0, 0.04, 1, 0.96])
+    out = RESULT_DIR / 'zoo_canonical_overview.png'
+    plt.savefig(out, dpi=150, bbox_inches='tight')
+    plt.close()
+    print(f'Saved: {out}')
+
+
+def visualize_cross_dataset_consistency():
+    """Plot 3: Heatmap showing canonical name consistency across datasets."""
+    # Collect canonical names for core body parts across datasets
+    core_parts = [
+        'pelvis', 'spine lower', 'spine mid', 'spine upper', 'neck', 'head',
+        'left collar', 'left upper arm', 'left forearm', 'left hand',
+        'right collar', 'right upper arm', 'right forearm', 'right hand',
+        'left upper leg', 'left lower leg', 'left foot', 'left toe',
+        'right upper leg', 'right lower leg', 'right foot', 'right toe',
+    ]
+
+    datasets = ['humanml3d', 'lafan1', '100style', 'bandai_namco', 'cmu_mocap', 'mixamo']
+    matrix = np.zeros((len(core_parts), len(datasets)), dtype=np.float32)
+
+    for j, ds in enumerate(datasets):
+        s = dict(np.load(f'data/processed/{ds}/skeleton.npz', allow_pickle=True))
+        canon_set = set(str(n) for n in s['canonical_names'])
+        for i, part in enumerate(core_parts):
+            matrix[i, j] = 1.0 if part in canon_set else 0.0
+
+    fig, ax = plt.subplots(figsize=(10, 12))
+    im = ax.imshow(matrix, cmap='RdYlGn', aspect='auto', vmin=0, vmax=1)
+
+    ax.set_xticks(range(len(datasets)))
+    ax.set_xticklabels(datasets, rotation=45, ha='right', fontsize=9)
+    ax.set_yticks(range(len(core_parts)))
+    ax.set_yticklabels(core_parts, fontsize=9)
+
+    # Annotate cells
+    for i in range(len(core_parts)):
+        for j in range(len(datasets)):
+            text = '✓' if matrix[i, j] > 0.5 else '✗'
+            color = 'white' if matrix[i, j] > 0.5 else 'red'
+            ax.text(j, i, text, ha='center', va='center', fontsize=10, color=color)
+
+    ax.set_title('Cross-Dataset Canonical Name Coverage\n(22 core human body parts)',
+                fontsize=12, fontweight='bold')
+    plt.colorbar(im, ax=ax, label='Present in dataset', shrink=0.6)
+    plt.tight_layout()
+    out = RESULT_DIR / 'canonical_consistency_heatmap.png'
+    plt.savefig(out, dpi=150, bbox_inches='tight')
+    plt.close()
+    print(f'Saved: {out}')
+
+
+def visualize_side_tag_stats():
+    """Plot 4: Bar chart of side tag distribution per dataset."""
+    all_data = []
+    labels = []
+
+    datasets = ['humanml3d', 'lafan1', '100style', 'bandai_namco', 'cmu_mocap', 'mixamo']
+    for ds in datasets:
+        s = dict(np.load(f'data/processed/{ds}/skeleton.npz', allow_pickle=True))
+        sg = SkeletonGraph.from_dict(s)
+        n_l = sum(1 for t in sg.side_tags if t == 'left')
+        n_r = sum(1 for t in sg.side_tags if t == 'right')
+        n_c = sum(1 for t in sg.side_tags if t == 'center')
+        all_data.append((n_l, n_r, n_c, len(sg.symmetry_pairs)))
+        labels.append(f'{ds}\n({sg.num_joints}j)')
+
+    # Add Zoo species
+    for sp in ['Dog', 'Cat', 'Horse', 'Eagle', 'Trex', 'Spider', 'Anaconda', 'Dragon']:
+        s = dict(np.load(f'data/processed/truebones_zoo/skeletons/{sp}.npz', allow_pickle=True))
+        sg = SkeletonGraph.from_dict(s)
+        n_l = sum(1 for t in sg.side_tags if t == 'left')
+        n_r = sum(1 for t in sg.side_tags if t == 'right')
+        n_c = sum(1 for t in sg.side_tags if t == 'center')
+        all_data.append((n_l, n_r, n_c, len(sg.symmetry_pairs)))
+        labels.append(f'{sp}\n({sg.num_joints}j)')
+
+    fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(16, 10))
+
+    x = range(len(labels))
+    lefts = [d[0] for d in all_data]
+    rights = [d[1] for d in all_data]
+    centers = [d[2] for d in all_data]
+    syms = [d[3] for d in all_data]
+
+    ax1.bar(x, lefts, color=SIDE_COLORS['left'], label='Left', alpha=0.8)
+    ax1.bar(x, rights, bottom=lefts, color=SIDE_COLORS['right'], label='Right', alpha=0.8)
+    ax1.bar(x, centers, bottom=[l+r for l,r in zip(lefts, rights)],
+           color=SIDE_COLORS['center'], label='Center', alpha=0.8)
+    ax1.set_xticks(x)
+    ax1.set_xticklabels(labels, fontsize=8)
+    ax1.set_ylabel('Joint Count')
+    ax1.set_title('Side Tag Distribution per Dataset/Species', fontweight='bold')
+    ax1.legend()
+    ax1.axvline(x=5.5, color='black', linestyle='--', alpha=0.3)
+    ax1.text(2.5, max(lefts)*2.5, 'Human', ha='center', fontsize=10, style='italic')
+    ax1.text(9.5, max(lefts)*2.5, 'Zoo Animals', ha='center', fontsize=10, style='italic')
+
+    ax2.bar(x, syms, color='#2ecc71', alpha=0.8)
+    ax2.set_xticks(x)
+    ax2.set_xticklabels(labels, fontsize=8)
+    ax2.set_ylabel('Symmetry Pairs')
+    ax2.set_title('Detected Symmetry Pairs per Dataset/Species', fontweight='bold')
+    ax2.axvline(x=5.5, color='black', linestyle='--', alpha=0.3)
+
+    plt.tight_layout()
+    out = RESULT_DIR / 'side_tag_stats.png'
+    plt.savefig(out, dpi=150, bbox_inches='tight')
+    plt.close()
+    print(f'Saved: {out}')
+
+
+if __name__ == '__main__':
+    print('Generating canonical name visualizations...\n')
+    visualize_human_datasets()
+    visualize_zoo_animals()
+    visualize_cross_dataset_consistency()
+    visualize_side_tag_stats()
+    print(f'\nAll saved to: {RESULT_DIR}/')