File size: 10,124 Bytes
d821aa4 | 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 | #!/usr/bin/env python3
"""
Script to update task_index in parquet files based on caption annotations.
This script:
1. Reads all caption files from the captions/ directory
2. Extracts unique captions and assigns them task_index values
3. Updates the tasks.jsonl file with the new tasks
4. Updates all parquet files to set the correct task_index based on frame_index
5. Adds subtask_end_frame_index column to indicate the end frame of each subtask
"""
import json
import os
from pathlib import Path
from collections import defaultdict
try:
import pyarrow.parquet as pq
import pyarrow as pa
except ImportError:
print("Error: pyarrow is not installed. Please install it with: pip install pyarrow")
print("Or activate the appropriate conda environment.")
exit(1)
def load_all_captions(captions_dir):
"""Load all caption files and return a mapping of episode_index -> segments."""
captions_dir = Path(captions_dir)
episode_captions = {}
# Get all caption files
caption_files = sorted(captions_dir.glob("episode_*.json"))
for caption_file in caption_files:
# Extract episode index from filename
episode_idx = int(caption_file.stem.replace("episode_", ""))
# Load caption data
with open(caption_file, 'r') as f:
segments = json.load(f)
episode_captions[episode_idx] = segments
print(f"Loaded {len(segments)} segments from {caption_file.name}")
return episode_captions
def extract_unique_tasks(episode_captions):
"""Extract all unique captions and create task_index mapping."""
unique_captions = set()
for episode_idx, segments in episode_captions.items():
for segment in segments:
unique_captions.add(segment['caption'])
# Sort captions for consistent ordering
sorted_captions = sorted(unique_captions)
# Create mapping: caption -> task_index
caption_to_task_index = {caption: idx for idx, caption in enumerate(sorted_captions)}
print(f"\nFound {len(sorted_captions)} unique tasks:")
for caption, idx in caption_to_task_index.items():
print(f" {idx}: {caption}")
return caption_to_task_index, sorted_captions
def update_tasks_file(tasks_file, tasks):
"""Update the tasks.jsonl file with new tasks."""
tasks_file = Path(tasks_file)
# Backup original file
backup_file = tasks_file.with_suffix('.jsonl.backup')
if tasks_file.exists():
import shutil
shutil.copy(tasks_file, backup_file)
print(f"\nBacked up original tasks.jsonl to {backup_file}")
# Write new tasks
with open(tasks_file, 'w') as f:
for task_index, task in enumerate(tasks):
task_entry = {"task_index": task_index, "task": task}
f.write(json.dumps(task_entry) + '\n')
print(f"Updated {tasks_file} with {len(tasks)} tasks")
def get_task_index_for_frame(frame_index, segments, caption_to_task_index):
"""Given a frame_index, find which segment it belongs to and return the task_index."""
for segment in segments:
if segment['start_frame'] <= frame_index < segment['end_frame']:
return caption_to_task_index[segment['caption']]
# If frame is beyond all segments, use the last segment's task
if segments and frame_index >= segments[-1]['end_frame']:
return caption_to_task_index[segments[-1]['caption']]
# If frame is before all segments (shouldn't happen), use the first segment's task
if segments and frame_index < segments[0]['start_frame']:
return caption_to_task_index[segments[0]['caption']]
return None
def get_subtask_end_frame_for_frame(frame_index, segments):
"""Given a frame_index, find which segment it belongs to and return the end_frame of that subtask."""
for segment in segments:
if segment['start_frame'] <= frame_index < segment['end_frame']:
return segment['end_frame']
# If frame is beyond all segments, use the last segment's end_frame
if segments and frame_index >= segments[-1]['end_frame']:
return segments[-1]['end_frame']
# If frame is before all segments (shouldn't happen), use the first segment's end_frame
if segments and frame_index < segments[0]['start_frame']:
return segments[0]['end_frame']
return None
def update_parquet_file(parquet_file, episode_idx, episode_captions, caption_to_task_index):
"""Update task_index and add subtask_end_frame_index in a single parquet file."""
# Read the parquet file
table = pq.read_table(parquet_file)
# Convert to dictionary for easier manipulation
data = {col: table.column(col).to_pylist() for col in table.column_names}
# Get the segments for this episode
if episode_idx not in episode_captions:
print(f"Warning: No caption data for episode {episode_idx}")
return False
segments = episode_captions[episode_idx]
# Update task_index and compute subtask_end_frame_index for each row based on frame_index
new_task_indices = []
new_subtask_end_frames = []
for frame_idx in data['frame_index']:
# Get task_index
task_idx = get_task_index_for_frame(frame_idx, segments, caption_to_task_index)
if task_idx is None:
print(f"Warning: Could not determine task_index for frame {frame_idx} in episode {episode_idx}")
task_idx = 0 # Default to 0 if we can't determine
new_task_indices.append(task_idx)
# Get subtask_end_frame_index
subtask_end_frame = get_subtask_end_frame_for_frame(frame_idx, segments)
if subtask_end_frame is None:
print(f"Warning: Could not determine subtask_end_frame_index for frame {frame_idx} in episode {episode_idx}")
subtask_end_frame = frame_idx # Default to current frame if we can't determine
new_subtask_end_frames.append(subtask_end_frame)
# Replace task_index column
data['task_index'] = new_task_indices
# Add or update subtask_end_frame_index column
data['subtask_end_frame_index'] = new_subtask_end_frames
# Create new table with updated data (preserve column order, add new column at the end if it didn't exist)
column_names = list(table.column_names)
if 'subtask_end_frame_index' not in column_names:
column_names.append('subtask_end_frame_index')
arrays = [pa.array(data[col]) for col in column_names]
new_table = pa.Table.from_arrays(arrays, names=column_names)
# Write back to parquet file
pq.write_table(new_table, parquet_file)
return True
def update_all_parquet_files(data_dir, episode_captions, caption_to_task_index):
"""Update task_index and add subtask_end_frame_index in all parquet files."""
data_dir = Path(data_dir)
# Find all parquet files
parquet_files = sorted(data_dir.rglob("episode_*.parquet"))
print(f"\nUpdating {len(parquet_files)} parquet files...")
for i, parquet_file in enumerate(parquet_files):
# Extract episode index from filename
episode_idx = int(parquet_file.stem.replace("episode_", ""))
print(f"Processing {i+1}/{len(parquet_files)}: {parquet_file.name}...", end=' ')
success = update_parquet_file(parquet_file, episode_idx, episode_captions, caption_to_task_index)
if success:
print("✓")
else:
print("✗")
print("\nAll parquet files updated!")
def update_info_file(info_file, num_tasks):
"""Update meta/info.json: ensure subtask_end_frame_index feature exists and total_tasks is correct."""
info_file = Path(info_file)
if not info_file.exists():
print(f"Warning: {info_file} does not exist, skipping info.json update")
return
# Backup original file
backup_file = info_file.with_suffix('.json.backup')
import shutil
shutil.copy(info_file, backup_file)
print(f"Backed up original info.json to {backup_file}")
with open(info_file, 'r') as f:
info = json.load(f)
# Update total_tasks
info['total_tasks'] = num_tasks
# Ensure features.subtask_end_frame_index is declared
features = info.setdefault('features', {})
features['subtask_end_frame_index'] = {
"dtype": "int64",
"shape": [1],
"names": None,
}
with open(info_file, 'w') as f:
json.dump(info, f, indent=4)
print(f"Updated {info_file} (total_tasks={num_tasks}, added subtask_end_frame_index feature)")
def main(dataset_path):
dataset_path = Path(dataset_path)
captions_dir = dataset_path / "captions"
tasks_file = dataset_path / "meta" / "tasks.jsonl"
info_file = dataset_path / "meta" / "info.json"
data_dir = dataset_path / "data"
print("=" * 80)
print(f"Task Index Update Script — {dataset_path}")
print("=" * 80)
# Step 1: Load all captions
print("\nStep 1: Loading caption files...")
episode_captions = load_all_captions(captions_dir)
# Step 2: Extract unique tasks
print("\nStep 2: Extracting unique tasks...")
caption_to_task_index, unique_tasks = extract_unique_tasks(episode_captions)
# Step 3: Update tasks.jsonl
print("\nStep 3: Updating tasks.jsonl...")
update_tasks_file(tasks_file, unique_tasks)
# Step 4: Update all parquet files
print("\nStep 4: Updating parquet files...")
update_all_parquet_files(data_dir, episode_captions, caption_to_task_index)
# Step 5: Update info.json
print("\nStep 5: Updating info.json...")
update_info_file(info_file, len(unique_tasks))
print("\n" + "=" * 80)
print("Task index update completed successfully!")
print("=" * 80)
if __name__ == "__main__":
import sys
if len(sys.argv) != 2:
print("Usage: python update_task_index.py <dataset_path>")
print("Example: python update_task_index.py /nobackup/zy_zhang/data/coffee_make_vla/r1lite-20260422-lerobot_trimmed")
sys.exit(1)
main(sys.argv[1])
|