File size: 7,539 Bytes
aaf8747 | 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 | #!/usr/bin/env python3
"""Prepare a portable clip-sized wm-proxy session from extracted metadata.
Only stdlib is needed. This prepares inputs; rendering needs the bundled compiler.
Recorded observations are copied unchanged. Output timestamps use the clip grid.
"""
import argparse
import copy
import json
from pathlib import Path
import sqlite3
from inspect_frame import resolve_source, rows
def dump(path, value):
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(json.dumps(value, ensure_ascii=False, indent=2) + '\n')
def dump_rows(path, values):
path.parent.mkdir(parents=True, exist_ok=True)
with path.open('x', encoding='utf-8') as stream:
for value in values:
stream.write(json.dumps(value, ensure_ascii=False, separators=(',', ':')) + '\n')
def prepare(index, source_root, alignment_file, output):
if output.exists():
raise FileExistsError(f'Choose a new output directory: {output}')
alignment = json.loads(alignment_file.read_text())
count = len(alignment['grid_ns'])
if count != 124 or any(len(alignment[s]) != count for s in ('left', 'right')):
raise ValueError('Expected two views with 124 source frame references')
conn = sqlite3.connect(index.resolve().as_uri() + '?mode=ro', uri=True)
conn.row_factory = sqlite3.Row
source_frames = {}
source_segments = {}
episode_roots = set()
episodes = set()
for side in ('left', 'right'):
source_frames[side] = []
for item in alignment[side]:
row = conn.execute('SELECT * FROM frames WHERE frame_id=?', (item['frame_id'],)).fetchone()
if row is None:
raise ValueError(f'Missing frame {item["frame_id"]}')
frame = dict(row)
if frame['status'] != 'estimated_alignment':
raise ValueError('Only usable aligned frames can be prepared')
source_frames[side].append(frame)
sid = frame['segment_id']
if sid not in source_segments:
seg = dict(conn.execute('SELECT * FROM segments WHERE segment_id=?', (sid,)).fetchone())
segment_dir = resolve_source(source_root, seg['output_directory'])
# Source files are in proxy decode order, regardless of legacy frame_index quirks.
raw = list(rows(segment_dir / 'source_frames.jsonl'))
aligned = list(rows(segment_dir / 'source_alignment.jsonl'))
if len(raw) != seg['frame_count'] or len(aligned) != seg['frame_count']:
raise ValueError(f'Source frame count mismatch for segment {sid}')
source_segments[sid] = (seg, raw, aligned)
episode_roots.add(segment_dir.parents[3])
episodes.add(seg['episode_id'])
conn.close()
if len(episode_roots) != 1 or len(episodes) != 1:
raise ValueError('A clip must belong to exactly one episode')
episode_root = episode_roots.pop()
all_frames = source_frames['left'] + source_frames['right']
domains = {f['server_clock_domain'] for f in all_frames}
if len(domains) != 1 or None in domains:
raise ValueError('Expected a single known server clock domain')
domain = domains.pop()
low = min(f['server_time_estimate_ns'] for f in all_frames) - 4_000_000_000
high = max(f['server_time_estimate_ns'] for f in all_frames) + 600_000_000
required_ticks = {f['world_capture_tick'] for f in all_frames}
world_rows = [r for r in rows(episode_root / '.spool/world_state.jsonl')
if r.get('server_clock_domain') == domain
and (low <= r.get('server_monotonic_ns', -1) <= high
or r.get('sample_seq') in required_ticks)]
if not required_ticks <= {r['sample_seq'] for r in world_rows}:
raise ValueError('Required world snapshots missing; extract the complete episode metadata')
clients = {}
agents = []
for side in ('left', 'right'):
agent_ids = {source_segments[f['segment_id']][0]['agent_id'] for f in source_frames[side]}
if len(agent_ids) != 1:
raise ValueError(f'Inconsistent camera ownership for {side}')
agent = agent_ids.pop()
agents.append(agent)
selected = []
for record in rows(episode_root / '.spool/agents' / agent / 'client.jsonl'):
if record.get('server_clock_domain') != domain:
continue
timestamp = record.get('server_time_estimate_ns')
if timestamp is None:
timestamp = record.get('server_broadcast_monotonic_ns')
if (timestamp is not None and low <= timestamp <= high) or record.get('sample_seq_estimate') in required_ticks:
selected.append(record)
if not selected:
raise ValueError(f'No context observations for {agent}')
clients[agent] = selected
output.mkdir(parents=True)
manifest = {'schema_version': '0.1.0', 'episode_id': episodes.pop(),
'agent_ids': agents, 'video_fps': 24, 'resolution': [1280, 720],
'preserve_decoded_rgb_timing': True,
'alignment_scope': 'release_clip_original_source_frames_on_24fps_output_grid',
'preparation_note': 'Recorded states unchanged; new visualization input, not bit-identical replay.'}
dump(output / 'manifest.json', manifest)
dump_rows(output / '.spool/world_state.jsonl', world_rows)
for side, agent in zip(('left', 'right'), agents):
raw_output, aligned_output = [], []
for i, frame in enumerate(source_frames[side]):
_, raw, aligned = source_segments[frame['segment_id']]
pts = alignment['grid_ns'][i] - alignment['grid_ns'][0]
for source, destination in [(raw, raw_output), (aligned, aligned_output)]:
record = copy.deepcopy(source[frame['proxy_frame_index']])
record.update(frame_index=i, video_pts_ns=pts, rgb_encoded_pts_ns=pts,
server_time_estimate_ns=frame['server_time_estimate_ns'],
server_clock_domain=frame['server_clock_domain'],
estimated_sample_seq=frame['world_capture_tick'],
release_source_frame_id=frame['frame_id'])
destination.append(record)
folder = output / 'agents' / agent
dump_rows(folder / 'frame_timestamps.jsonl', raw_output)
dump_rows(folder / 'frame_timestamps.aligned.jsonl', aligned_output)
dump_rows(output / '.spool/agents' / agent / 'client.jsonl', clients[agent])
summary = {'num_frames_per_view': count, 'agent_ids': agents, 'world_records': len(world_rows),
'client_records': {key: len(value) for key, value in clients.items()},
'context_before_seconds': 4, 'context_after_seconds': 0.6,
'server_clock_domain': domain, 'raw_observations_modified': False}
dump(output / 'preparation.json', summary)
return summary
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--index', type=Path, required=True)
parser.add_argument('--source-root', type=Path, required=True)
parser.add_argument('--alignment', type=Path, required=True)
parser.add_argument('--output', type=Path, required=True)
args = parser.parse_args()
print(json.dumps(prepare(args.index, args.source_root, args.alignment, args.output), indent=2))
if __name__ == '__main__':
main()
|