File size: 36,212 Bytes
b89dd15 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 | #!/usr/bin/env python3
"""
HO3D 模型规范化工具
统一坐标系约定,为多视角渲染做准备
支持 PyTorch DataLoader 多线程并行处理
"""
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader
import trimesh
import os
import time
from scipy.spatial.transform import Rotation
from pathlib import Path
from typing import Optional, List, Dict, Any
from tqdm import tqdm
class HO3DModelNormalizer:
def __init__(self, output_coordinate_system: str = "z_up_finger_z"):
"""HO3D/Contact 模型标准化器
内部分析仍按 Y-up 约定进行:
+Y: 上方向 (Up) - 手腕到指尖
+Z: 前方向 (Forward) - 手掌法向量
+X: 右方向 (Right) - 拇指到小指
参数:
output_coordinate_system:
- "y_up" : 最终 OBJ 仍保持 Y-up
- "z_up_finger_z": 最终 OBJ 为 Z-up, 手指沿 +Z (推荐给 InstantMesh)
"""
self.output_coordinate_system = output_coordinate_system
self.coordinate_convention = {
'up_axis': 'Y', # 手腕到指尖方向
'forward_axis': 'Z', # 手掌朝向
'right_axis': 'X' # 拇指到小指方向
}
def load_mesh(self, obj_path):
"""加载OBJ模型"""
try:
mesh = trimesh.load(obj_path)
if hasattr(mesh, 'vertices'):
return mesh
else:
# 处理多个mesh的情况
combined_vertices = []
combined_faces = []
face_offset = 0
for m in mesh.geometry.values():
combined_vertices.append(m.vertices)
combined_faces.append(m.faces + face_offset)
face_offset += len(m.vertices)
vertices = np.vstack(combined_vertices)
faces = np.vstack(combined_faces)
return trimesh.Trimesh(vertices=vertices, faces=faces)
except Exception as e:
print(f"加载模型失败 {obj_path}: {e}")
return None
def analyze_hand_orientation(self, hand_vertices):
"""
分析手部模型的主要朝向
返回主轴方向向量
"""
# 计算边界框
bbox_min = hand_vertices.min(axis=0)
bbox_max = hand_vertices.max(axis=0)
bbox_size = bbox_max - bbox_min
# 使用PCA找到主要方向
centered_vertices = hand_vertices - hand_vertices.mean(axis=0)
cov_matrix = np.cov(centered_vertices.T)
eigenvalues, eigenvectors = np.linalg.eig(cov_matrix)
# 按特征值排序
sorted_indices = np.argsort(eigenvalues)[::-1]
principal_axes = eigenvectors[:, sorted_indices]
return {
'bbox_size': bbox_size,
'center': hand_vertices.mean(axis=0),
'principal_axes': principal_axes,
'eigenvalues': eigenvalues[sorted_indices]
}
def detect_hand_palm_normal(self, hand_vertices, hand_faces):
"""
检测手掌法向量方向
"""
mesh = trimesh.Trimesh(vertices=hand_vertices, faces=hand_faces)
# 计算面法向量
face_normals = mesh.face_normals
face_areas = mesh.area_faces
# 加权平均法向量(按面积加权)
weighted_normal = np.average(face_normals, axis=0, weights=face_areas)
weighted_normal = weighted_normal / np.linalg.norm(weighted_normal)
return weighted_normal
def compute_wrist_to_fingertip_direction(self, hand_vertices):
"""
计算手腕到指尖的方向向量
"""
# 使用手部顶点的边界框来估计手腕到指尖方向
# 假设手部顶点按照MANO模型的顺序排列
# 方法1: 使用主成分分析
centered_vertices = hand_vertices - hand_vertices.mean(axis=0)
cov_matrix = np.cov(centered_vertices.T)
eigenvalues, eigenvectors = np.linalg.eigh(cov_matrix)
# 最大特征值对应的特征向量通常是手的主要延伸方向
main_direction = eigenvectors[:, -1] # 最大特征值对应的特征向量
# 方法2: 使用几何分析 - 找到最远的两个点
distances = np.linalg.norm(hand_vertices[:, None] - hand_vertices[None, :], axis=2)
max_dist_indices = np.unravel_index(np.argmax(distances), distances.shape)
point1, point2 = hand_vertices[max_dist_indices[0]], hand_vertices[max_dist_indices[1]]
geometric_direction = point2 - point1
geometric_direction = geometric_direction / np.linalg.norm(geometric_direction)
# 选择与主成分方向一致的方向
if np.dot(main_direction, geometric_direction) < 0:
main_direction = -main_direction
return main_direction
def validate_coordinate_system(self, target_x, target_y, target_z, hand_vertices):
"""
验证并调整坐标系,确保符合约定
+Y: 上方向 (手腕到指尖) - 主要延伸方向
+Z: 前方向 (手掌法向量)
+X: 右方向 (拇指到小指)
"""
# 检查是否为右手坐标系 Y × Z = X
cross_product = np.cross(target_y, target_z)
if np.dot(cross_product, target_x) < 0:
# 如果不是右手坐标系,翻转X轴
target_x = -target_x
print("⚠️ 调整X轴方向以确保右手坐标系")
# 验证Y轴是否为主要延伸方向
hand_center = hand_vertices.mean(axis=0)
# 计算沿各轴的投影范围
x_projections = np.dot(hand_vertices - hand_center, target_x)
y_projections = np.dot(hand_vertices - hand_center, target_y)
z_projections = np.dot(hand_vertices - hand_center, target_z)
x_range = np.max(x_projections) - np.min(x_projections)
y_range = np.max(y_projections) - np.min(y_projections)
z_range = np.max(z_projections) - np.min(z_projections)
print(f"📏 各轴投影范围:")
print(f" X轴范围: {x_range:.4f}")
print(f" Y轴范围: {y_range:.4f}")
print(f" Z轴范围: {z_range:.4f}")
# Y轴应该是最大的延伸方向
if y_range < max(x_range, z_range):
print("⚠️ 警告: Y轴不是最大延伸方向,可能需要调整")
else:
print("✅ Y轴确实是主要延伸方向")
print(f"\\n✅ 坐标系验证完成:")
print(f" X轴 (拇指到小指): {target_x}")
print(f" Y轴 (手腕到指尖): {target_y}")
print(f" Z轴 (手掌法向量): {target_z}")
return target_x, target_y, target_z
def compute_normalization_transform(self, hand_vertices, hand_faces, object_vertices=None):
"""
计算规范化变换矩阵
严格按照坐标系约定: +Y=Up(手腕到指尖), +Z=Forward(手掌法向量), +X=Right(拇指到小指)
"""
# 分析手部朝向
hand_analysis = self.analyze_hand_orientation(hand_vertices)
# 检测手掌法向量
palm_normal = self.detect_hand_palm_normal(hand_vertices, hand_faces)
# 重新构建目标坐标系 - 根据用户反馈调整
# 用户反馈: 前应该是左,后应该是右,上下是对的
# 这意味着我们需要重新映射坐标轴
# 首先确定Y轴: 手腕到指尖方向 (主要延伸方向) - 保持不变
target_y = self.compute_wrist_to_fingertip_direction(hand_vertices)
target_y = target_y / np.linalg.norm(target_y)
# 重新定义Z轴: 原来的X轴方向 (这样前后会变成左右)
# 先计算原来的手掌法向量
palm_normal_orig = palm_normal / np.linalg.norm(palm_normal)
# 计算原来的X轴方向
temp_z = palm_normal_orig - np.dot(palm_normal_orig, target_y) * target_y
temp_z = temp_z / np.linalg.norm(temp_z)
temp_x = np.cross(target_y, temp_z)
temp_x = temp_x / np.linalg.norm(temp_x)
# 新的坐标系: 交换X和Z的角色
target_z = temp_x # 原来的X轴变成新的Z轴
target_x = -temp_z # 原来的Z轴的负方向变成新的X轴
print(f"🔄 重新映射坐标轴: 前后↔左右")
# 验证并调整方向,确保符合约定
target_x, target_y, target_z = self.validate_coordinate_system(
target_x, target_y, target_z, hand_vertices
)
# 构建旋转矩阵
# 目标坐标系的轴向量作为行,构成旋转矩阵
# 这样变换后,原来的X轴变成target_x方向,Y轴变成target_y方向,Z轴变成target_z方向
rotation_matrix = np.row_stack([target_x, target_y, target_z])
# 计算平移(移到原点)
if object_vertices is not None:
all_vertices = np.vstack([hand_vertices, object_vertices])
else:
all_vertices = hand_vertices
center = all_vertices.mean(axis=0)
# 构建4x4变换矩阵
transform_matrix = np.eye(4)
transform_matrix[:3, :3] = rotation_matrix
transform_matrix[:3, 3] = -rotation_matrix @ center # 先旋转再平移
# 重新构建target_axes用于返回
target_axes = np.column_stack([target_x, target_y, target_z])
return {
'transform_matrix': transform_matrix,
'rotation_matrix': rotation_matrix,
'translation': -center,
'center': center,
'target_axes': target_axes,
'palm_normal': palm_normal,
'analysis': hand_analysis
}
def apply_transform(self, vertices, transform_matrix):
"""应用变换矩阵到顶点"""
# 转换为齐次坐标
vertices_homo = np.hstack([vertices, np.ones((len(vertices), 1))])
# 应用变换
transformed_homo = (transform_matrix @ vertices_homo.T).T
# 返回3D坐标
return transformed_homo[:, :3]
def normalize_combined_model(self, hand_obj_path, output_dir, verbose=True):
"""
规范化组合模型(手+物体)
参数:
hand_obj_path: 输入模型路径
output_dir: 输出目录
verbose: 是否打印详细日志(默认True,并行处理时设为False)
"""
if verbose:
print(f"正在规范化模型: {hand_obj_path}")
# 加载组合模型
combined_mesh = self.load_mesh(hand_obj_path)
if combined_mesh is None:
return None
# 解析手部和物体部分(基于OBJ文件的组结构)
hand_vertices, hand_faces, object_vertices, object_faces = self.parse_combined_obj(hand_obj_path)
if hand_vertices is None:
print("无法解析手部和物体部分")
return None
# 计算规范化变换
transform_info = self.compute_normalization_transform(
hand_vertices, hand_faces, object_vertices
)
# 应用变换
normalized_hand = self.apply_transform(hand_vertices, transform_info['transform_matrix'])
normalized_object = self.apply_transform(object_vertices, transform_info['transform_matrix'])
# 保存规范化后的模型
output_path = Path(output_dir) / "normalized_hand_object_combined.obj"
self.save_normalized_obj(
normalized_hand, hand_faces,
normalized_object, object_faces,
output_path, transform_info,
verbose=verbose
)
return {
'output_path': output_path,
'transform_info': transform_info,
'original_vertices': {
'hand': hand_vertices,
'object': object_vertices
},
'normalized_vertices': {
'hand': normalized_hand,
'object': normalized_object
}
}
def parse_combined_obj(self, obj_path):
"""
解析组合OBJ文件,分离手部和物体
基于固定的顶点数量分割:前778个顶点是手部,其余是物体
"""
try:
# 使用trimesh加载完整模型
mesh = trimesh.load(obj_path)
vertices = mesh.vertices
faces = mesh.faces
print(f"📊 加载模型: {len(vertices)}个顶点, {len(faces)}个面片")
# 基于固定数量分割
hand_vertex_count = 778
if len(vertices) < hand_vertex_count:
print(f"❌ 顶点数量不足: {len(vertices)} < {hand_vertex_count}")
return None, None, None, None
# 分离顶点
hand_vertices = vertices[:hand_vertex_count]
object_vertices = vertices[hand_vertex_count:]
# 分离面片
hand_faces = []
object_faces = []
for face in faces:
# 检查面片的所有顶点是否都属于手部
if np.all(face < hand_vertex_count):
hand_faces.append(face)
# 检查面片的所有顶点是否都属于物体
elif np.all(face >= hand_vertex_count):
# 调整物体面片的索引,使其从0开始
adjusted_face = face - hand_vertex_count
object_faces.append(adjusted_face)
# 跨越手部和物体的面片(通常不应该存在)
else:
print(f"⚠️ 发现跨越手部和物体的面片: {face}")
hand_faces = np.array(hand_faces) if hand_faces else None
object_faces = np.array(object_faces) if object_faces else None
print(f"✅ 分离完成:")
print(f" 手部: {len(hand_vertices)}个顶点, {len(hand_faces) if hand_faces is not None else 0}个面片")
print(f" 物体: {len(object_vertices)}个顶点, {len(object_faces) if object_faces is not None else 0}个面片")
return hand_vertices, hand_faces, object_vertices, object_faces
except Exception as e:
print(f"❌ 解析OBJ文件失败: {e}")
import traceback
traceback.print_exc()
return None, None, None, None
def save_normalized_obj(
self,
hand_vertices,
hand_faces,
object_vertices,
object_faces,
output_path,
transform_info,
verbose=True,
):
"""保存规范化后的 OBJ 文件
说明:
- 只在磁盘上保存最终标准化后的 OBJ
- 不再写入 .transform.npz 等中间变换信息文件
"""
os.makedirs(os.path.dirname(output_path), exist_ok=True)
# 根据配置决定最终导出的坐标系
coord = getattr(self, "output_coordinate_system", "y_up")
hand_vertices_out = hand_vertices
object_vertices_out = object_vertices
if coord == "z_up_finger_z":
# 在内部 Y-up 标准化的基础上,再旋转到 Z-up 且手指朝向 +Z
# 这里使用和渲染侧一致的约定:
# X' = X
# Y' = -Z
# Z' = Y
R = np.array(
[
[1.0, 0.0, 0.0],
[0.0, 0.0, -1.0],
[0.0, 1.0, 0.0],
],
dtype=np.float32,
)
hand_vertices_out = hand_vertices_out @ R.T
object_vertices_out = object_vertices_out @ R.T
with open(output_path, "w") as f:
# 写入头部信息
f.write("# Normalized HO3D Hand-Object Model\n")
if coord == "z_up_finger_z":
f.write("# Coordinate Convention: Z-up, fingers along +Z (InstantMesh)\n")
else:
f.write("# Coordinate Convention: +Y=Up, +Z=Forward, +X=Right\n")
f.write("# Transform applied for multi-view rendering\n\n")
# 写入手部顶点
f.write("# Hand vertices\n")
for vertex in hand_vertices_out:
f.write(f"v {vertex[0]:.6f} {vertex[1]:.6f} {vertex[2]:.6f}\n")
# 写入物体顶点
f.write("\n# Object vertices\n")
for vertex in object_vertices_out:
f.write(f"v {vertex[0]:.6f} {vertex[1]:.6f} {vertex[2]:.6f}\n")
# 写入手部面片
f.write("\n# Hand faces\n")
f.write("g hand\n")
for face in hand_faces:
f.write(f"f {face[0] + 1} {face[1] + 1} {face[2] + 1}\n")
# 写入物体面片
f.write("\n# Object faces\n")
f.write("g object\n")
vertex_offset = len(hand_vertices)
for face in object_faces:
f.write(
f"f {face[0] + vertex_offset + 1} "
f"{face[1] + vertex_offset + 1} "
f"{face[2] + vertex_offset + 1}\n"
)
# 不再保存 transform.npz,只在日志中提示 OBJ 路径
if verbose:
print(f"规范化模型已保存: {output_path}")
# ====================== DataLoader 并行处理实现 ======================
# Worker 进程级别的全局标准化器缓存
_WORKER_NORMALIZER_CACHE = {}
def _get_worker_normalizer(output_coordinate_system: str) -> HO3DModelNormalizer:
"""为每个 worker 进程懒加载一个 HO3DModelNormalizer 实例"""
global _WORKER_NORMALIZER_CACHE
if output_coordinate_system not in _WORKER_NORMALIZER_CACHE:
_WORKER_NORMALIZER_CACHE[output_coordinate_system] = HO3DModelNormalizer(
output_coordinate_system=output_coordinate_system
)
return _WORKER_NORMALIZER_CACHE[output_coordinate_system]
def _process_single_mesh(task: dict, output_coordinate_system: str) -> dict:
"""在 DataLoader worker 中执行的单个网格标准化逻辑
参数:
task: 包含 sequence, frame_id, input_mesh, output_obj 的字典
output_coordinate_system: 输出坐标系
返回:
包含处理结果的字典
"""
import tempfile
import shutil
seq = task["sequence"]
frame_id = task["frame_id"]
input_mesh = task["input_mesh"]
output_obj = task["output_obj"]
output_dir = os.path.dirname(output_obj)
# 已存在的文件直接跳过(不需要先创建目录)
if os.path.exists(output_obj):
return {
**task,
"success": True,
"skipped": True,
"error": None,
}
# 使用临时目录避免多进程冲突
temp_dir = None
try:
# 创建唯一的临时目录
temp_dir = tempfile.mkdtemp(prefix=f"normalize_{seq}_{frame_id}_")
normalizer = _get_worker_normalizer(output_coordinate_system)
# 并行处理时禁用详细日志,避免输出混乱
result = normalizer.normalize_combined_model(input_mesh, temp_dir, verbose=False)
if not result:
return {
**task,
"success": False,
"skipped": False,
"error": "normalize_combined_model returned None",
}
default_output = os.path.join(temp_dir, "normalized_hand_object_combined.obj")
if not os.path.exists(default_output):
return {
**task,
"success": False,
"skipped": False,
"error": "normalized_hand_object_combined.obj not found in temp dir",
}
# 成功后才创建输出目录,避免产生空目录
os.makedirs(output_dir, exist_ok=True)
# 移动到目标位置
shutil.move(default_output, output_obj)
return {
**task,
"success": True,
"skipped": False,
"error": None,
}
except Exception as e:
return {
**task,
"success": False,
"skipped": False,
"error": str(e),
}
finally:
# 清理临时目录
if temp_dir and os.path.exists(temp_dir):
try:
shutil.rmtree(temp_dir)
except Exception:
pass
class NormalizeTaskDataset(Dataset):
"""用于 DataLoader 的标准化任务 Dataset
每个 __getitem__ 会在 worker 进程中执行一个网格的标准化。
"""
def __init__(self, tasks: List[dict], output_coordinate_system: str):
self.tasks = tasks
self.output_coordinate_system = output_coordinate_system
def __len__(self):
return len(self.tasks)
def __getitem__(self, idx):
task = self.tasks[idx]
return _process_single_mesh(task, self.output_coordinate_system)
def build_tasks_from_directory(
input_dir: str,
output_dir: str,
max_sequences: Optional[int] = None,
max_frames_per_sequence: Optional[int] = None,
) -> List[dict]:
"""从输入目录构建所有需要处理的任务列表
参数:
input_dir: 输入目录,包含多个序列子目录
output_dir: 输出目录
max_sequences: 最大序列数(用于测试)
max_frames_per_sequence: 每个序列最大帧数(用于测试)
返回:
任务列表,每个任务包含 sequence, frame_id, input_mesh, output_obj
"""
input_path = Path(input_dir)
output_path = Path(output_dir)
if not input_path.exists():
print(f"❌ 输入目录不存在: {input_dir}")
return []
tasks = []
sequences = sorted([d for d in input_path.iterdir() if d.is_dir()])
if max_sequences is not None:
sequences = sequences[:max_sequences]
for seq_dir in sequences:
seq_name = seq_dir.name
# 查找该序列下的所有 .ply 和 .obj 文件
mesh_files = list(seq_dir.glob("*.ply")) + list(seq_dir.glob("*.obj"))
mesh_files = sorted(mesh_files)
if max_frames_per_sequence is not None:
mesh_files = mesh_files[:max_frames_per_sequence]
# 创建输出目录
seq_output_dir = output_path / seq_name
seq_output_dir.mkdir(parents=True, exist_ok=True)
for mesh_file in mesh_files:
frame_id = mesh_file.stem # 文件名(不含扩展名)
output_obj = seq_output_dir / f"normalized_{frame_id}.obj"
tasks.append({
"sequence": seq_name,
"frame_id": frame_id,
"input_mesh": str(mesh_file),
"output_obj": str(output_obj),
})
return tasks
def parallel_normalize(
input_dir: str,
output_dir: str,
output_coordinate_system: str = "z_up_finger_z",
num_workers: int = 4,
max_sequences: Optional[int] = None,
max_frames_per_sequence: Optional[int] = None,
) -> bool:
"""使用 PyTorch DataLoader 并行标准化
参数:
input_dir: 输入目录
output_dir: 输出目录
output_coordinate_system: 输出坐标系
num_workers: DataLoader worker 数量(CPU 进程数)
max_sequences: 最大序列数(用于测试)
max_frames_per_sequence: 每个序列最大帧数(用于测试)
返回:
是否成功处理至少一个文件
"""
print("=" * 60)
print("🚀 DataLoader 并行标准化模式")
print(f"📁 输入目录: {input_dir}")
print(f"📁 输出目录: {output_dir}")
print(f"📐 输出坐标系: {output_coordinate_system}")
print(f"🔧 并行进程数: {num_workers}")
if max_sequences:
print(f"🧪 测试模式: 最多处理 {max_sequences} 个序列")
if max_frames_per_sequence:
print(f"🧪 测试模式: 每序列最多 {max_frames_per_sequence} 帧")
print("=" * 60)
# 构建任务列表
tasks = build_tasks_from_directory(
input_dir=input_dir,
output_dir=output_dir,
max_sequences=max_sequences,
max_frames_per_sequence=max_frames_per_sequence,
)
if not tasks:
print("❌ 没有找到可处理的网格文件")
return False
print(f"📦 共构建 {len(tasks)} 个任务")
# 创建 Dataset 和 DataLoader
dataset = NormalizeTaskDataset(tasks, output_coordinate_system)
# 自定义 collate_fn 直接返回结果列表
def collate_fn(batch):
return batch
dataloader = DataLoader(
dataset,
batch_size=1,
shuffle=False,
num_workers=num_workers,
collate_fn=collate_fn,
prefetch_factor=2 if num_workers > 0 else None,
)
# 统计计数器
success_count = 0
skipped_count = 0
failed_count = 0
failed_tasks = []
start_time = time.time()
# 使用 tqdm 显示进度条
for batch_results in tqdm(dataloader, desc="处理进度", unit="帧"):
for result in batch_results:
if result["success"]:
if result["skipped"]:
skipped_count += 1
else:
success_count += 1
else:
failed_count += 1
failed_tasks.append({
"sequence": result["sequence"],
"frame_id": result["frame_id"],
"error": result["error"],
})
elapsed_time = time.time() - start_time
# 打印统计结果
print("\n" + "=" * 60)
print("📊 并行标准化完成!")
print(f"✅ 成功处理: {success_count} 个")
print(f"⏭️ 跳过已存在: {skipped_count} 个")
print(f"❌ 失败: {failed_count} 个")
print(f"⏱️ 总耗时: {elapsed_time:.1f} 秒")
if success_count + skipped_count > 0:
avg_time = elapsed_time / (success_count + skipped_count + failed_count)
print(f"📈 平均速度: {avg_time:.3f} 秒/帧")
if failed_tasks and len(failed_tasks) <= 10:
print("\n❌ 失败任务详情:")
for ft in failed_tasks:
print(f" - {ft['sequence']}/{ft['frame_id']}: {ft['error']}")
elif failed_tasks:
print(f"\n❌ 失败任务太多,只显示前10个:")
for ft in failed_tasks[:10]:
print(f" - {ft['sequence']}/{ft['frame_id']}: {ft['error']}")
print("=" * 60)
return success_count > 0 or skipped_count > 0
def main():
"""批量规范化所有重建结果 - 支持多数据集和并行处理"""
import argparse
parser = argparse.ArgumentParser(
description="HO3D 手-物体模型标准化工具(支持多线程并行处理)",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
示例用法:
# 单个文件标准化
python model_normalizer.py --input path/to/file.obj --output path/to/output_dir
# 目录批量并行处理(推荐!)
python model_normalizer.py --input /path/to/input_dir --output /path/to/output_dir --parallel --workers 8
# 测试模式(每序列处理3帧,共处理5个序列)
python model_normalizer.py --input dir --output out --parallel --workers 4 --max-sequences 5 --max-frames 3
# 批量处理(使用配置文件)
python model_normalizer.py --batch
# 指定输出坐标系
python model_normalizer.py --input input.obj --output out/ --coord z_up_finger_z
""",
)
parser.add_argument(
"--input", "-i", type=str, help="输入文件/目录路径"
)
parser.add_argument(
"--output", "-o", type=str, help="输出目录路径"
)
parser.add_argument(
"--parallel", "-p",
action="store_true",
help="启用 DataLoader 多线程并行处理(用于目录批量处理)",
)
parser.add_argument(
"--workers", "-w",
type=int,
default=4,
help="并行处理的 worker 数量(默认: 4)",
)
parser.add_argument(
"--max-sequences",
type=int,
default=None,
help="最大处理序列数(用于测试)",
)
parser.add_argument(
"--max-frames",
type=int,
default=None,
help="每序列最大处理帧数(用于测试)",
)
parser.add_argument(
"--batch",
action="store_true",
help="批量处理模式(使用 simple_multiview_config.py 配置)",
)
parser.add_argument(
"--coord",
type=str,
default=None,
choices=["y_up", "z_up_finger_z"],
help="输出坐标系(默认: z_up_finger_z)",
)
args = parser.parse_args()
# 导入配置
try:
import sys as _sys
import os as _os
# 确保可以从 render 根目录导入 simple_multiview_config
render_root = _os.path.dirname(_os.path.dirname(__file__))
if render_root not in _sys.path:
_sys.path.insert(0, render_root)
from simple_multiview_config import (
get_dataset_config,
PROCESSING_CONFIG,
NORMALIZATION_OUTPUT_CONFIG,
)
dataset_config = get_dataset_config()
config_coord = NORMALIZATION_OUTPUT_CONFIG.get("coordinate_system", "z_up_finger_z")
except ImportError as e:
# 向后兼容:如果没有新配置,使用默认设置
print(f"⚠️ 无法导入配置: {e}")
dataset_config = {
"input_dir": "ho3d_reconstruction_results",
"output_dir": "ho3d_reconstruction_results",
"input_file": "hand_object_combined.obj",
"output_subdir": "clean_combined",
"normalized_subdir": "normalized",
}
config_coord = "z_up_finger_z"
print("⚠️ 使用默认配置 (向后兼容模式)")
# 确定最终使用的坐标系
coord = args.coord if args.coord else config_coord
# 并行目录批量处理模式(推荐)
if args.input and args.output and args.parallel:
input_path = Path(args.input)
if input_path.is_dir():
success = parallel_normalize(
input_dir=args.input,
output_dir=args.output,
output_coordinate_system=coord,
num_workers=args.workers,
max_sequences=args.max_sequences,
max_frames_per_sequence=args.max_frames,
)
return 0 if success else 1
else:
print("❌ 并行模式需要 --input 为目录路径")
return 1
# 单文件模式
elif args.input and args.output:
input_path = Path(args.input)
# 如果 input 是目录,使用单线程批量处理
if input_path.is_dir():
print("=" * 60)
print("🔧 目录批量标准化模式(单线程)")
print(f"📁 输入目录: {args.input}")
print(f"📁 输出目录: {args.output}")
print(f"📐 输出坐标系: {coord}")
print("💡 提示: 使用 --parallel 启用多线程加速")
print("=" * 60)
normalizer = HO3DModelNormalizer(output_coordinate_system=coord)
output_path = Path(args.output)
success_count = 0
fail_count = 0
sequences = sorted([d for d in input_path.iterdir() if d.is_dir()])
for seq_dir in sequences:
mesh_files = list(seq_dir.glob("*.ply")) + list(seq_dir.glob("*.obj"))
for mesh_file in sorted(mesh_files):
seq_output_dir = output_path / seq_dir.name
seq_output_dir.mkdir(parents=True, exist_ok=True)
output_obj = seq_output_dir / f"normalized_{mesh_file.stem}.obj"
if output_obj.exists():
print(f"⏭️ 跳过: {output_obj}")
success_count += 1
continue
print(f"处理: {mesh_file}")
result = normalizer.normalize_combined_model(
str(mesh_file), str(seq_output_dir)
)
if result:
# 重命名输出文件
default_out = seq_output_dir / "normalized_hand_object_combined.obj"
if default_out.exists():
default_out.rename(output_obj)
success_count += 1
else:
fail_count += 1
print("\n" + "=" * 60)
print(f"📊 批量处理完成: 成功 {success_count} 个,失败 {fail_count} 个")
print("=" * 60)
return 0 if fail_count == 0 else 1
# 单文件处理
else:
print("=" * 60)
print("🔧 单文件标准化模式")
print(f"📄 输入文件: {args.input}")
print(f"📁 输出目录: {args.output}")
print(f"📐 输出坐标系: {coord}")
print("=" * 60)
normalizer = HO3DModelNormalizer(output_coordinate_system=coord)
result = normalizer.normalize_combined_model(args.input, args.output)
if result:
print(f"\n✅ 标准化完成!")
print(f" 输出文件: {result['output_path']}")
return 0
else:
print(f"\n❌ 标准化失败")
return 1
# 批量处理模式(使用配置文件)
elif args.batch or (not args.input and not args.output):
print("=" * 60)
print("🔧 批量标准化模式(配置文件)")
print(f"📂 当前数据集: {PROCESSING_CONFIG.get('active_dataset', 'N/A')}")
print(f"📁 输入目录: {dataset_config['input_dir']}")
print(f"📁 输出目录: {dataset_config['output_dir']}")
print(f"📐 输出坐标系: {coord}")
print("=" * 60)
normalizer = HO3DModelNormalizer(output_coordinate_system=coord)
# 处理所有重建结果
input_dir = Path(dataset_config["input_dir"])
output_base_dir = Path(dataset_config["output_dir"])
if not input_dir.exists():
print(f"❌ 输入目录不存在: {input_dir}")
return 1
success_count = 0
fail_count = 0
for sequence_dir in input_dir.iterdir():
if sequence_dir.is_dir():
combined_obj = sequence_dir / dataset_config["input_file"]
if combined_obj.exists():
print(f"\n处理序列: {sequence_dir.name}")
# 创建输出目录结构
output_sequence_dir = (
output_base_dir
/ dataset_config["output_subdir"]
/ sequence_dir.name
)
normalized_dir = (
output_sequence_dir / dataset_config["normalized_subdir"]
)
normalized_dir.mkdir(parents=True, exist_ok=True)
# 执行规范化
result = normalizer.normalize_combined_model(
str(combined_obj), str(normalized_dir)
)
if result:
print(f"✅ 规范化完成: {sequence_dir.name}")
success_count += 1
else:
print(f"❌ 规范化失败: {sequence_dir.name}")
fail_count += 1
print("\n" + "=" * 60)
print(f"📊 批量处理完成: 成功 {success_count} 个,失败 {fail_count} 个")
print("=" * 60)
return 0 if fail_count == 0 else 1
else:
parser.print_help()
print("\n❌ 错误: 请指定 --input 和 --output,或使用 --batch 模式")
return 1
if __name__ == "__main__":
import sys
sys.exit(main()) |