ScanQA_voxel_data / preprocessing /export_sampled_frames.py
Yezixiao's picture
Upload folder using huggingface_hub
84d92e1 verified
"""
1. 导出的核心数据类型
对于每一个被采样到的视频帧,脚本会提取并保存以下 5 种数据:
Color (RGB 图像): 从原始流中解压 JPEG 颜色数据,支持根据 --image_size 参数重新调整分辨率(Resize),最终保存为 .jpg 格式。
Depth (深度图): 解压深度数据,并将其转化为 16-bit 灰度图的 .png 格式保存。
Pose (相机外参/位姿): 获取该帧对应的相机到世界坐标系的 4x4 变换矩阵(Camera-to-World)。特别地,脚本会读取 ScanNet 场景的元数据(scene_id.txt),获取 axisAlignment 矩阵,将原始的相机位姿对齐到统一的全局重力坐标系下,最后保存为 .txt 文件。
Instance Mask (实例分割掩码): 如果同级目录下存在 2d-instance-filt.zip 文件,脚本会自动去解压并匹配对应帧的 2D 实例/语义标签掩码,保存为 .png。
Intrinsics (相机内参): 解析元数据中的焦距和主点坐标(fx, fy, mx, my),构建 4x4 的深度图相机内参矩阵并保存。
"""
import argparse
import os
import sys
import cv2 # Requires opencv-python
import numpy as np
from tqdm import tqdm # Requires tqdm
import glob
import concurrent.futures
import math
import png # Requires pypng
import zipfile # Added for zip file handling
import shutil # Added for file copying
sensor_dir = os.path.dirname(os.path.abspath(__file__))
sys.path.append(sensor_dir)
# Assuming SensorData is in the same directory or PYTHONPATH is set
from SensorData import SensorData
# Helper function to parse scene_id.txt
def parse_scene_meta_file(filename):
"""Parses the scene_id.txt file to extract axis alignment and depth intrinsics."""
metadata = {}
try:
with open(filename, 'r') as f:
for line in f:
parts = line.strip().split(' = ')
if len(parts) == 2:
key, value = parts
metadata[key.strip()] = value.strip()
# Extract and reshape axis alignment matrix
axis_align_str = metadata.get('axisAlignment')
if axis_align_str:
axis_align_vals = list(map(float, axis_align_str.split()))
if len(axis_align_vals) == 16:
axis_align_matrix = np.array(axis_align_vals).reshape(4, 4)
else:
print(f"Warning: Invalid number of values for axisAlignment in {filename}. Expected 16, got {len(axis_align_vals)}.")
axis_align_matrix = None
else:
print(f"Warning: axisAlignment key not found in {filename}.")
axis_align_matrix = None
# Construct depth intrinsics matrix
try:
fx = float(metadata['fx_depth'])
fy = float(metadata['fy_depth'])
mx = float(metadata['mx_depth'])
my = float(metadata['my_depth'])
depth_intrinsics = np.array([
[fx, 0, mx, 0],
[0, fy, my, 0],
[0, 0, 1, 0],
[0, 0, 0, 1]
])
except KeyError as e:
print(f"Warning: Missing depth intrinsic key {e} in {filename}.")
depth_intrinsics = None
except ValueError as e:
print(f"Warning: Invalid value for depth intrinsic key in {filename}: {e}.")
depth_intrinsics = None
return axis_align_matrix, depth_intrinsics
except FileNotFoundError:
print(f"Warning: Scene metadata file not found: {filename}")
return None, None
except Exception as e:
print(f"Warning: Error parsing scene metadata file {filename}: {e}")
return None, None
def save_matrix_to_file(matrix, filename):
"""Saves a numpy matrix to a text file."""
with open(filename, 'w') as f:
for row in matrix:
f.write(' '.join(map(str, row)) + '\n')
def export_scene_sampled_frames(sens_file_path, output_base_dir, num_frames_to_sample, split, image_size=None, min_valid_components_per_frame_initial=0, min_valid_frames_per_scene=1):
"""Exports uniformly sampled pose, depth, color, instance mask, and intrinsics for a single scene, applying axis alignment to poses."""
scene_id = os.path.basename(os.path.dirname(sens_file_path))
scene_dir_path = os.path.dirname(sens_file_path)
print(f"Processing scene: {scene_id} (split: {split})")
print(f'Loading {sens_file_path}...')
try:
sd = SensorData(sens_file_path)
except NameError:
print("Error: SensorData class failed to be imported.")
return None
except Exception as e:
print(f"Error loading SensorData from {sens_file_path}: {e}")
return None
print(f'Loaded {len(sd.frames)} frames.')
if not hasattr(sd, 'frames') or not sd.frames:
print("Error: SensorData object does not contain frames or is empty.")
return None
total_raw_frames = len(sd.frames)
if total_raw_frames == 0:
print(f"Scene {scene_id} has 0 frames. Skipping.")
return scene_id
# --- Locate additional input files ---
scene_meta_path = os.path.join(scene_dir_path, f"{scene_id}.txt")
instance_zip_path = os.path.join(scene_dir_path, f"{scene_id}_2d-instance-filt.zip")
if not os.path.exists(instance_zip_path):
instance_zip_path = os.path.join(scene_dir_path, f"{scene_id}_2d-instance.zip")
instance_zip = None
if os.path.exists(instance_zip_path):
try:
instance_zip = zipfile.ZipFile(instance_zip_path, 'r')
print(f"Opened instance mask zip: {instance_zip_path}")
except zipfile.BadZipFile:
print(f"Warning: Bad zip file: {instance_zip_path}. Instance masks will be unavailable.")
instance_zip = None
except Exception as e:
print(f"Warning: Error opening zip file {instance_zip_path}: {e}. Instance masks will be unavailable.")
instance_zip = None
else:
print(f"Warning: Instance mask zip file not found at {instance_zip_path} or fallback. Instance masks will be unavailable.")
# --- Phase 1: Validate all available frames to create a list of candidates ---
print(f"Validating all {total_raw_frames} available frames for scene {scene_id}...")
valid_frame_candidates_info = []
for original_idx in tqdm(range(total_raw_frames), desc=f"Validating raw frames for {scene_id}", leave=False):
try:
frame = sd.frames[original_idx]
num_available_components = 0
if hasattr(frame, 'camera_to_world') and frame.camera_to_world is not None:
num_available_components += 1
if hasattr(frame, 'decompress_depth') and hasattr(sd, 'depth_compression_type') and sd.depth_compression_type.lower() != 'unknown':
num_available_components += 1
if hasattr(frame, 'decompress_color') and hasattr(sd, 'color_compression_type') and sd.color_compression_type.lower() == 'jpeg':
num_available_components += 1
instance_mask_potentially_available = False
if instance_zip:
mask_filename_in_zip = f'instance-filt/{original_idx}.png'
try:
instance_zip.getinfo(mask_filename_in_zip)
instance_mask_potentially_available = True
except KeyError:
instance_mask_potentially_available = False
except Exception:
instance_mask_potentially_available = False
if instance_mask_potentially_available:
num_available_components +=1
if num_available_components >= min_valid_components_per_frame_initial:
valid_frame_candidates_info.append({"original_idx": original_idx})
except IndexError:
print(f"Warning: Raw frame index {original_idx} out of bounds during validation for scene {scene_id}.")
continue
except Exception as e_val:
print(f"Warning: Error validating raw frame {original_idx} for scene {scene_id}: {e_val}. Skipping candidate.")
continue
if not valid_frame_candidates_info:
print(f"No valid frame candidates found for scene {scene_id} after initial validation. Skipping scene processing.")
if instance_zip: instance_zip.close()
return None
num_candidates = len(valid_frame_candidates_info)
print(f"Found {num_candidates} valid frame candidates for scene {scene_id}.")
# --- Sampling Logic based on validated candidates ---
if num_candidates <= num_frames_to_sample:
selected_candidate_indices = np.arange(num_candidates)
else:
selected_candidate_indices = np.linspace(0, num_candidates - 1, num_frames_to_sample, dtype=int)
selected_candidate_indices = np.unique(selected_candidate_indices)
actual_indices_to_process = [valid_frame_candidates_info[i]["original_idx"] for i in selected_candidate_indices]
if not actual_indices_to_process: # Should not happen if valid_frame_candidates_info is not empty
print(f"No frames selected for processing for scene {scene_id} after sampling (num_candidates: {num_candidates}, num_frames_to_sample: {num_frames_to_sample}). Skipping.")
if instance_zip: instance_zip.close()
return None
print(f"Selected {len(actual_indices_to_process)} frames for export. First few original indices: {actual_indices_to_process[:10]}...")
# --- Create output directories ---
pose_output_dir = os.path.join(output_base_dir, 'pose', split, scene_id)
depth_output_dir = os.path.join(output_base_dir, 'depth', split, scene_id)
color_output_dir = os.path.join(output_base_dir, 'color', split, scene_id)
instance_output_dir = os.path.join(output_base_dir, 'instance', split, scene_id)
intrinsic_base_output_dir = os.path.join(output_base_dir, 'intrinsic', split)
os.makedirs(pose_output_dir, exist_ok=True)
os.makedirs(depth_output_dir, exist_ok=True)
os.makedirs(color_output_dir, exist_ok=True)
os.makedirs(instance_output_dir, exist_ok=True)
os.makedirs(intrinsic_base_output_dir, exist_ok=True)
# --- Process scene-level data ---
axis_align_matrix, intrinsics_matrix = parse_scene_meta_file(scene_meta_path)
if intrinsics_matrix is not None:
intrinsic_out_filename = os.path.join(intrinsic_base_output_dir, f'intrinsic_depth_{scene_id}.txt')
save_matrix_to_file(intrinsics_matrix, intrinsic_out_filename)
print(f"Saved intrinsics to {intrinsic_out_filename}")
else:
print(f"Warning: Could not read/parse or save intrinsics for scene {scene_id} from {scene_meta_path}.")
if axis_align_matrix is None:
print(f"Warning: Could not read/parse axis alignment matrix for scene {scene_id} from {scene_meta_path}. Poses will NOT be aligned.")
# --- Phase 2: Process and Save Selected Frames ---
processed_attempt_count = 0
final_valid_frames_count = 0
total_skipped_components_in_export = 0
min_successful_components_per_frame_export = min_valid_components_per_frame_initial
try:
for original_idx in tqdm(actual_indices_to_process, desc=f"Exporting selected frames for {scene_id}", leave=False):
try:
frame = sd.frames[original_idx]
components_skipped_this_frame_export = 0
pose_saved_path = None
depth_saved_path = None
color_saved_path = None
instance_saved_path = None
color_export_failed_critically = False
# Export Pose
if hasattr(frame, 'camera_to_world') and frame.camera_to_world is not None:
pose_matrix_original = frame.camera_to_world
pose_filename = os.path.join(pose_output_dir, f'{original_idx:06d}.txt')
if axis_align_matrix is not None:
try:
aligned_pose = np.dot(axis_align_matrix, pose_matrix_original)
save_matrix_to_file(aligned_pose, pose_filename)
pose_saved_path = pose_filename
except ValueError as e:
print(f"Warning: Error applying axis alignment for frame {original_idx}: {e}. Saving original.")
save_matrix_to_file(pose_matrix_original, pose_filename)
pose_saved_path = pose_filename
except Exception as e:
print(f"Warning: Unexpected error during pose alignment/saving for frame {original_idx}: {e}. Skip pose comp.")
components_skipped_this_frame_export += 1
else:
save_matrix_to_file(pose_matrix_original, pose_filename)
pose_saved_path = pose_filename
else:
print(f"Warning: Pose data not found for pre-validated frame {original_idx} during export. Skip pose comp.")
components_skipped_this_frame_export += 1
# Export Depth
if hasattr(frame, 'decompress_depth') and hasattr(sd, 'depth_compression_type') and sd.depth_compression_type.lower() != 'unknown':
depth_data = frame.decompress_depth(sd.depth_compression_type)
if depth_data is not None:
try:
depth_image = np.frombuffer(depth_data, dtype=np.uint16).reshape(sd.depth_height, sd.depth_width)
depth_filename = os.path.join(depth_output_dir, f'{original_idx:06d}.png')
depth_image_reshaped = depth_image.reshape(-1, depth_image.shape[1]).tolist()
with open(depth_filename, 'wb') as f_png:
writer = png.Writer(width=depth_image.shape[1], height=depth_image.shape[0], bitdepth=16, greyscale=True)
writer.write(f_png, depth_image_reshaped)
depth_saved_path = depth_filename
except (ValueError, AttributeError, Exception) as e: # Catch specific and general errors
print(f"Warning: Error processing/writing depth for frame {original_idx}: {e}. Skip depth comp.")
components_skipped_this_frame_export += 1
else:
print(f"Warning: Decompressed null depth data for frame {original_idx}. Skip depth comp.")
components_skipped_this_frame_export += 1
else:
print(f"Warning: Depth data/decompression not available for pre-validated frame {original_idx}. Skip depth comp.")
components_skipped_this_frame_export += 1
# Export Color
if hasattr(frame, 'decompress_color') and hasattr(sd, 'color_compression_type') and sd.color_compression_type.lower() == 'jpeg':
color_image = frame.decompress_color(sd.color_compression_type)
if color_image is not None:
try:
color_filename = os.path.join(color_output_dir, f'{original_idx:06d}.jpg')
if image_size:
color_image_resized = cv2.resize(color_image, (image_size[1], image_size[0]))
else:
color_image_resized = color_image
color_image_bgr = cv2.cvtColor(color_image_resized, cv2.COLOR_RGB2BGR)
if not cv2.imwrite(color_filename, color_image_bgr):
raise ValueError(f"cv2.imwrite failed for color {color_filename}")
color_saved_path = color_filename
except Exception as e:
print(f"Warning: Error saving color image for frame {original_idx}: {e}. CRITICAL: Skip color & FRAME.")
components_skipped_this_frame_export += 1
color_export_failed_critically = True
else:
print(f"Warning: Decompressed null color data for frame {original_idx}. CRITICAL: Skip color & FRAME.")
components_skipped_this_frame_export += 1
color_export_failed_critically = True
else:
print(f"Warning: Color data/decompression not JPEG for pre-validated frame {original_idx}. CRITICAL: Skip color & FRAME.")
components_skipped_this_frame_export += 1
color_export_failed_critically = True
if color_export_failed_critically:
if pose_saved_path and os.path.exists(pose_saved_path): os.remove(pose_saved_path)
if depth_saved_path and os.path.exists(depth_saved_path): os.remove(depth_saved_path)
total_skipped_components_in_export += (4 - components_skipped_this_frame_export) # Add remaining potential skips
processed_attempt_count += 1
continue
# Export Instance Mask
if instance_zip:
mask_filename_in_zip = f'instance-filt/{original_idx}.png'
instance_output_filename = os.path.join(instance_output_dir, f'{original_idx:06d}.png')
try:
with instance_zip.open(mask_filename_in_zip, 'r') as mask_file:
mask_data = mask_file.read()
instance_mask = cv2.imdecode(np.frombuffer(mask_data, np.uint8), cv2.IMREAD_UNCHANGED)
if instance_mask is None: raise ValueError(f"cv2.imdecode failed for {mask_filename_in_zip}")
if image_size:
target_h, target_w = image_size
instance_mask = cv2.resize(instance_mask, (target_w, target_h), interpolation=cv2.INTER_NEAREST)
if not cv2.imwrite(instance_output_filename, instance_mask):
raise ValueError(f"cv2.imwrite failed for instance {instance_output_filename}")
instance_saved_path = instance_output_filename
except KeyError:
print(f"Warning: Instance mask {mask_filename_in_zip} not in zip for frame {original_idx}. Skip instance comp.")
components_skipped_this_frame_export += 1
except Exception as e:
print(f"Warning: Error processing instance mask {mask_filename_in_zip} for frame {original_idx}: {e}. Skip instance comp.")
components_skipped_this_frame_export += 1
elif not instance_zip: # Instance masks were not available for the scene
pass # Not a skip for this frame if unavailable for scene, handled by num_successfully_exported
processed_attempt_count += 1
total_skipped_components_in_export += components_skipped_this_frame_export
num_successfully_exported_components = 0
if pose_saved_path: num_successfully_exported_components +=1
if depth_saved_path: num_successfully_exported_components +=1
if color_saved_path: num_successfully_exported_components +=1
if instance_saved_path: num_successfully_exported_components +=1
if num_successfully_exported_components >= min_successful_components_per_frame_export:
final_valid_frames_count += 1
else:
print(f"Info: Frame {original_idx} did not meet min successful components ({num_successfully_exported_components}/{min_successful_components_per_frame_export}). Cleaning up.")
if pose_saved_path and os.path.exists(pose_saved_path): os.remove(pose_saved_path)
if depth_saved_path and os.path.exists(depth_saved_path): os.remove(depth_saved_path)
if color_saved_path and os.path.exists(color_saved_path): os.remove(color_saved_path)
if instance_saved_path and os.path.exists(instance_saved_path): os.remove(instance_saved_path)
except IndexError:
print(f"Warning: Frame index {original_idx} out of bounds for scene {scene_id} during export. Critical error.")
break
except Exception as e:
print(f"Warning: Unhandled error processing selected frame {original_idx} for scene {scene_id}: {e}. Skipping frame.")
total_skipped_components_in_export += 4
processed_attempt_count += 1
continue
finally:
if instance_zip:
instance_zip.close()
print(f"Closed instance mask zip for {scene_id}")
scene_export_successful = False
if total_raw_frames == 0:
scene_export_successful = True
elif processed_attempt_count > 0 and final_valid_frames_count >= min_valid_frames_per_scene:
scene_export_successful = True
elif not actual_indices_to_process and total_raw_frames > 0 :
scene_export_successful = False
else:
scene_export_successful = False
if scene_export_successful:
print(f'Finished exporting for scene {scene_id} (split: {split}).')
print(f'Initial valid candidates: {num_candidates}. Selected for processing: {len(actual_indices_to_process)}.')
print(f'Attempted export for {processed_attempt_count} selected frames.')
print(f'Number of finally valid frames (>= {min_successful_components_per_frame_export} components exported): {final_valid_frames_count} (required >= {min_valid_frames_per_scene}).')
if total_skipped_components_in_export > 0:
print(f'Skipped {total_skipped_components_in_export} components during export phase.')
return scene_id
else:
print(f"Failed to export scene {scene_id} (split: {split}).")
print(f" Initial valid candidates: {num_candidates}. Selected for processing: {len(actual_indices_to_process)}.")
print(f" Attempted export for {processed_attempt_count} selected frames.")
print(f" Number of finally valid frames (>= {min_successful_components_per_frame_export} components exported): {final_valid_frames_count} (required >= {min_valid_frames_per_scene}).")
if not valid_frame_candidates_info and total_raw_frames > 0:
print(" Reason: No frame candidates passed initial validation.")
elif not actual_indices_to_process and valid_frame_candidates_info:
print(" Reason: No frames were selected from candidates for processing.")
elif final_valid_frames_count < min_valid_frames_per_scene and processed_attempt_count > 0 :
print(f" Reason: Number of finally valid frames ({final_valid_frames_count}) is less than the required minimum ({min_valid_frames_per_scene}).")
return None
def main():
parser = argparse.ArgumentParser(description="Export uniformly sampled pose (axis-aligned), depth, color, and instance masks from ScanNet .sens files, along with intrinsics, organised by train/val splits.")
parser.add_argument('--scans_dir', required=True, help='Path to the directory containing scene subdirectories (e.g., data/scannet/scans)')
parser.add_argument('--output_dir', required=True, help='Path to the base directory where output train/val folders (containing pose/depth/color/instance_mask/intrinsic) will be saved')
parser.add_argument('--train_val_splits_path', type=str, default=None, help='Path to the directory containing scannetv2_train.txt and scannetv2_val.txt (optional). If provided, scenes will be sorted into train/val subfolders.')
parser.add_argument('--num_frames', type=int, default=32, help='Number of frames to uniformly sample (default: 32)')
parser.add_argument('--max_workers', type=int, default=None, help='Maximum number of processes to use for parallel processing (default: number of cores)')
parser.add_argument('--image_size', type=int, nargs=2, metavar=('HEIGHT', 'WIDTH'), default=None, help='Target image size (height width) to resize color and instance images to (default: None)')
parser.add_argument('--skip_existing', action='store_true', help='Skip processing scenes if their ID is found in the corresponding successful_scenes_<split>.txt in the output directory, or if output data directories exist.')
parser.add_argument('--scene_list_file', type=str, default=None, help='Path to a text file containing a list of scene IDs to process, one per line. If provided, --scans_dir is still used to locate these scenes.')
parser.add_argument('--min_valid_components_per_frame', type=int, default=0, help='Minimum number of components (pose, depth, color, instance) that must be available for a frame to be a candidate for sampling, AND successfully exported for a frame to be considered valid post-export. Default 0.')
parser.add_argument('--min_valid_frames_per_scene', type=int, default=1, help='Minimum number of valid frames required for a scene to be considered successfully processed. Default 1.')
opt = parser.parse_args()
print("Script Options:")
print(vars(opt))
if opt.image_size and len(opt.image_size) != 2:
print("Error: --image_size requires two arguments: HEIGHT WIDTH")
sys.exit(1)
if opt.image_size:
print(f"Color images will be resized to Height={opt.image_size[0]}, Width={opt.image_size[1]}")
train_scenes = set()
val_scenes = set()
if opt.train_val_splits_path:
train_file_path = os.path.join(opt.train_val_splits_path, 'scannetv2_train.txt')
val_file_path = os.path.join(opt.train_val_splits_path, 'scannetv2_val.txt')
try:
with open(train_file_path, 'r') as f:
train_scenes = set(line.strip() for line in f if line.strip())
print(f"Loaded {len(train_scenes)} train scene IDs from {train_file_path}")
except FileNotFoundError:
print(f"Warning: Train split file not found: {train_file_path}.")
except Exception as e:
print(f"Warning: Error reading train split file {train_file_path}: {e}")
try:
with open(val_file_path, 'r') as f:
val_scenes = set(line.strip() for line in f if line.strip())
print(f"Loaded {len(val_scenes)} val scene IDs from {val_file_path}")
except FileNotFoundError:
print(f"Warning: Validation split file not found: {val_file_path}.")
except Exception as e:
print(f"Warning: Error reading validation split file {val_file_path}: {e}")
if not train_scenes and not val_scenes:
print("Warning: Train/Val split path provided, but failed to load any scenes from the split files.")
else:
print("Train/Val split path not provided. Scenes will not be assigned to train/val splits and processing might be skipped depending on skip logic.")
if not os.path.exists(opt.output_dir):
print(f"Creating output directory: {opt.output_dir}")
os.makedirs(opt.output_dir, exist_ok=True)
successful_scenes_list_files = {
'train': os.path.join(opt.output_dir, 'successful_scenes_train.txt'),
'val': os.path.join(opt.output_dir, 'successful_scenes_val.txt')
}
existing_successful_scenes = {'train': set(), 'val': set()}
if opt.skip_existing:
for split_key, list_file in successful_scenes_list_files.items(): # Use split_key consistently
if os.path.exists(list_file):
try:
with open(list_file, 'r') as f:
existing_successful_scenes[split_key] = set(line.strip() for line in f if line.strip())
print(f"Loaded {len(existing_successful_scenes[split_key])} previously successful {split_key} scene IDs from {list_file}")
except Exception as e:
print(f"Warning: Could not read {list_file}: {e}. Proceeding without skipping based on this list for {split_key} split.")
scenes_to_consider = []
if opt.scene_list_file:
if not os.path.exists(opt.scene_list_file):
print(f"Error: Scene list file not found: {opt.scene_list_file}")
sys.exit(1)
if not opt.scans_dir: # scans_dir is needed to locate the data
print(f"Error: --scans_dir must be provided when using --scene_list_file.")
sys.exit(1)
try:
with open(opt.scene_list_file, 'r') as f:
target_scene_ids = [line.strip() for line in f if line.strip()]
print(f"Loaded {len(target_scene_ids)} scene IDs from {opt.scene_list_file}")
for scene_id_val in target_scene_ids: # Use scene_id_val to avoid conflict
scene_dir_path = os.path.join(opt.scans_dir, scene_id_val)
if os.path.isdir(scene_dir_path):
scenes_to_consider.append(scene_dir_path)
else:
print(f"Warning: Scene directory for ID '{scene_id_val}' not found at {scene_dir_path}. Skipping.")
except Exception as e:
print(f"Error reading scene list file {opt.scene_list_file}: {e}")
sys.exit(1)
else:
if not opt.scans_dir:
print(f"Error: --scans_dir must be provided if --scene_list_file is not used.")
sys.exit(1)
try:
scenes_to_consider = [d.path for d in os.scandir(opt.scans_dir) if d.is_dir()]
except FileNotFoundError:
print(f"Error: Scans directory not found: {opt.scans_dir}")
sys.exit(1)
except Exception as e:
print(f"Error scanning directory {opt.scans_dir}: {e}")
sys.exit(1)
if not scenes_to_consider:
print(f"No scene directories found or specified to process. Exiting.")
sys.exit(0)
print(f"Found {len(scenes_to_consider)} potential scene directories to evaluate.")
total_scenes_found = len(scenes_to_consider)
successful_scenes_this_run = {'train': [], 'val': []}
scenes_to_process_args_list = []
skipped_due_to_sens_or_skip_flag = 0
skipped_due_to_no_split = 0
for scene_dir_path in scenes_to_consider:
current_scene_id = os.path.basename(scene_dir_path) # Renamed to avoid conflict
sens_file_path = os.path.join(scene_dir_path, f"{current_scene_id}.sens")
if not os.path.exists(sens_file_path):
print(f"Warning: .sens file not found for scene {current_scene_id}. Skipping.")
skipped_due_to_sens_or_skip_flag += 1
continue
current_split = None # Renamed to avoid conflict
if opt.train_val_splits_path:
if current_scene_id in train_scenes:
current_split = 'train'
elif current_scene_id in val_scenes:
current_split = 'val'
else:
skipped_due_to_no_split += 1
continue
else:
print(f"Warning: --train_val_splits_path not provided. Skipping scene {current_scene_id} as split assignment is required.")
skipped_due_to_no_split += 1
continue
if opt.skip_existing:
if current_scene_id in existing_successful_scenes[current_split]:
print(f"Scene {current_scene_id} (split: {current_split}) found in successful list. Skipping.")
skipped_due_to_sens_or_skip_flag += 1
continue
# Simplified output path check - check for one key output directory for the scene in its split
scene_output_check_path = os.path.join(opt.output_dir, current_split, 'pose', current_scene_id) # Check e.g. pose dir
if os.path.exists(scene_output_check_path):
print(f"Output data likely exists for scene {current_scene_id} (split: {current_split}). Skipping.")
skipped_due_to_sens_or_skip_flag += 1
continue
scenes_to_process_args_list.append((sens_file_path, opt.output_dir, opt.num_frames, current_split, opt.image_size, opt.min_valid_components_per_frame, opt.min_valid_frames_per_scene))
num_scenes_to_actually_process = len(scenes_to_process_args_list)
print(f"Skipped {skipped_due_to_sens_or_skip_flag} scenes based on missing .sens or --skip_existing flag.")
if skipped_due_to_no_split > 0:
print(f"Skipped {skipped_due_to_no_split} scenes because they were not found in train/val lists or --train_val_splits_path was not provided/required.")
print(f"Attempting to process {num_scenes_to_actually_process} scenes.")
if num_scenes_to_actually_process == 0:
print("No scenes left to process. Exiting.")
sys.exit(0)
print(f"Starting parallel export using up to {opt.max_workers or os.cpu_count()} workers...")
with concurrent.futures.ProcessPoolExecutor(max_workers=opt.max_workers) as executor:
futures_map = {}
# Unpack arguments correctly for submit
for sens_path_arg, output_base_arg, num_frames_arg, split_arg, image_size_arg, min_comp_arg, min_frames_arg in scenes_to_process_args_list:
# Need scene_id for futures_map key association, extract from sens_path_arg
proc_scene_id = os.path.basename(os.path.dirname(sens_path_arg))
future = executor.submit(export_scene_sampled_frames, sens_path_arg, output_base_arg, num_frames_arg, split_arg, image_size_arg, min_comp_arg, min_frames_arg)
futures_map[future] = (proc_scene_id, split_arg) # Use proc_scene_id and split_arg
processed_counts_this_run = {'train': 0, 'val': 0} # Renamed for clarity
failed_counts_this_run = {'train': 0, 'val': 0} # Renamed for clarity
for future in tqdm(concurrent.futures.as_completed(futures_map), total=len(futures_map), desc="Processing Scenes"):
result_scene_id, result_split = futures_map[future] # Get associated scene_id and split
try:
returned_scene_id = future.result()
if returned_scene_id is not None: # export_scene_sampled_frames returns scene_id on success
successful_scenes_this_run[result_split].append(returned_scene_id)
processed_counts_this_run[result_split] += 1
else:
failed_counts_this_run[result_split] += 1
except Exception as exc:
print(f'\nScene {result_scene_id} (split: {result_split}) generated an exception during execution: {exc}')
failed_counts_this_run[result_split] += 1
all_successful_scenes_combined = {'train': set(), 'val': set()} # Renamed for clarity
for split_key in ['train', 'val']: # Use split_key consistently
if successful_scenes_this_run[split_key]:
current_split_successful_combined = existing_successful_scenes[split_key].union(set(successful_scenes_this_run[split_key]))
all_successful_scenes_combined[split_key] = current_split_successful_combined
list_file = successful_scenes_list_files[split_key]
try:
with open(list_file, 'w') as f:
for scene_id_val in sorted(list(current_split_successful_combined)):
f.write(scene_id_val + '\n')
print(f"Updated successful {split_key} scenes list at: {list_file}")
except Exception as e:
print(f"Error writing successful {split_key} scenes list to {list_file}: {e}")
else:
all_successful_scenes_combined[split_key] = existing_successful_scenes[split_key]
total_processed_in_this_run = sum(processed_counts_this_run.values())
total_failed_in_this_run = sum(failed_counts_this_run.values())
total_successful_over_all_runs = sum(len(s) for s in all_successful_scenes_combined.values())
print("\n" + "=" * 30)
print("Processing Summary:")
print(f"Total potential scenes found: {total_scenes_found}")
print(f"Scenes skipped (missing .sens / --skip_existing): {skipped_due_to_sens_or_skip_flag}")
print(f"Scenes skipped (no split assignment): {skipped_due_to_no_split}")
print(f"Scenes submitted for processing in this run: {num_scenes_to_actually_process}")
print(f" - Train: {processed_counts_this_run['train']} successful, {failed_counts_this_run['train']} failed")
print(f" - Val: {processed_counts_this_run['val']} successful, {failed_counts_this_run['val']} failed")
print(f"Total successfully processed scenes (this run): {total_processed_in_this_run}")
print(f"Total failed/incomplete scenes (this run): {total_failed_in_this_run}")
print("-" * 30)
print(f"Total successful scenes across all runs:")
print(f" - Train: {len(all_successful_scenes_combined['train'])} (list: {successful_scenes_list_files['train']})")
print(f" - Val: {len(all_successful_scenes_combined['val'])} (list: {successful_scenes_list_files['val']})")
print(f" - Overall: {total_successful_over_all_runs}")
print("Note: Success indicates the scene processing completed. Check logs for skipped components or frames.")
print("=" * 30)
if __name__ == '__main__':
main()