File size: 13,357 Bytes
7d93608 | 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 | #!/usr/bin/env python3
"""Generate train samples for cups_cot with structured, parse-friendly CoT text."""
from __future__ import annotations
import argparse
import json
import os
import random
import re
from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor, as_completed
from pathlib import Path
from generators.cups_shuffle import (
CupsShuffleConfig,
_jitter_color as _cups_jitter_color,
_shuffle_keyframes as _cups_shuffle_keyframes,
generate_cups_shuffle,
)
PROJECT_ROOT = Path(__file__).resolve().parents[1]
ROOT = PROJECT_ROOT
CATEGORY = "cups_cot"
SEED_PATTERN = re.compile(r"cups_cot_train_s(\d+)\.mp4$")
def _label_from_index(index: int) -> str:
return chr(ord("A") + int(index))
def _default_id_from_rel_path(rel_path: str) -> str:
value = Path(rel_path).with_suffix("").as_posix().replace("/", "_").strip("_")
return value or "sample"
def _split_category(name: str) -> tuple[str, str]:
if name == "cups_cot":
return "cups_cot", "cups_cot"
if name.startswith("hidden_container_"):
return "hidden_container", name.replace("hidden_container_", "", 1)
if name == "random_dots":
return "biological", "random_dots"
if "_" in name:
major, sub = name.split("_", 1)
return major, sub
return name, name
def _make_cups_cfg(seed: int) -> CupsShuffleConfig:
return CupsShuffleConfig(
width=1280,
height=720,
fps=30,
intro_duration=2.0,
cover_duration=2.0,
shuffle_duration=16.0,
swaps=4,
seed=int(seed),
ball_radius=18,
cup_count=4,
)
def _consume_style_rng(cfg: CupsShuffleConfig, rng: random.Random) -> None:
if not cfg.randomize_style:
return
cfg.size_scale = rng.uniform(0.92, 1.08)
cfg.arc_scale = rng.uniform(0.9, 1.2)
cfg.lane_scale = rng.uniform(0.9, 1.15)
cfg.cup_color = _cups_jitter_color(cfg.cup_color, rng, 14)
cfg.cup_shade = _cups_jitter_color(cfg.cup_shade, rng, 12)
cfg.cup_outline = _cups_jitter_color(cfg.cup_outline, rng, 12)
cfg.cup_rim = _cups_jitter_color(cfg.cup_rim, rng, 14)
cfg.cup_inner = _cups_jitter_color(cfg.cup_inner, rng, 8)
cfg.cup_highlight = _cups_jitter_color(cfg.cup_highlight, rng, 16)
def _derive_trace(seed: int) -> dict:
cfg = _make_cups_cfg(seed)
rng = random.Random(cfg.seed)
_consume_style_rng(cfg, rng)
ball_cup_id = rng.randint(0, cfg.cup_count - 1)
keyframes = _cups_shuffle_keyframes(cfg, cfg.cup_count)
if len(keyframes) != cfg.swaps + 1:
raise RuntimeError(
f"Unexpected keyframe length for seed={seed}: "
f"{len(keyframes)} (expected {cfg.swaps + 1})"
)
swaps: list[tuple[str, str]] = []
for step in range(cfg.swaps):
prev_order = keyframes[step]
next_order = keyframes[step + 1]
diff_positions = [
idx
for idx, (prev_cup, next_cup) in enumerate(zip(prev_order, next_order))
if prev_cup != next_cup
]
if len(diff_positions) != 2:
raise RuntimeError(
f"Seed {seed} step {step + 1}: expected 2 changed positions, got "
f"{len(diff_positions)}"
)
left_label = _label_from_index(prev_order[diff_positions[0]])
right_label = _label_from_index(prev_order[diff_positions[1]])
cup_a, cup_b = sorted((left_label, right_label))
swaps.append((cup_a, cup_b))
final_order = keyframes[-1]
final_order_labels = [_label_from_index(cup_id) for cup_id in final_order]
final_pos = final_order.index(ball_cup_id)
answer = _label_from_index(final_pos)
return {
"ball_start_under": _label_from_index(ball_cup_id),
"swap_count": int(cfg.swaps),
"swaps": swaps,
"final_order": final_order_labels,
"ball_end_under": answer,
}
def _build_cot_text(trace: dict) -> str:
lines = [
"COT_VERSION:1",
"TASK:cups_cot",
"CUP_LABELS:A,B,C,D",
"LABEL_SEMANTICS:IDENTITY_FIXED",
"INIT_ORDER_LEFT_TO_RIGHT:A,B,C,D",
f"BALL_START_UNDER:{trace['ball_start_under']}",
f"SWAP_COUNT:{trace['swap_count']}",
]
for idx, (cup_a, cup_b) in enumerate(trace["swaps"], start=1):
lines.append(f"SWAP_{idx:02d}:{cup_a}<->{cup_b}")
lines.extend(
[
f"FINAL_ORDER_LEFT_TO_RIGHT:{','.join(trace['final_order'])}",
f"BALL_END_UNDER:{trace['ball_end_under']}",
f"**Answer:{trace['ball_end_under']}**",
]
)
return "\n".join(lines)
def _build_question() -> str:
return (
"Based on the video, track cup swaps step by step and determine which cup "
"covers the red ball at the end. Cups are labeled A, B, C, and D from left "
"to right at the start. Answer in the format **Answer:X**."
)
def _build_entry(rel_path: str, text: str, question: str, answer: str) -> dict:
major, sub = _split_category(CATEGORY)
abs_video = str((ROOT / rel_path).resolve())
return {
"id": _default_id_from_rel_path(rel_path),
"path": rel_path,
"category": CATEGORY,
"major": major,
"sub": sub,
"text": text,
"question": question,
"answer": answer,
"data_split": "train",
"video_path": abs_video,
"image_gt": "",
"image_paths": [],
}
def _load_dataset(path: Path) -> list[dict]:
if not path.exists():
return []
try:
raw = json.loads(path.read_text(encoding="utf-8"))
except json.JSONDecodeError:
return []
if isinstance(raw, dict) and isinstance(raw.get("items"), list):
raw = raw["items"]
if not isinstance(raw, list):
return []
return [dict(item) for item in raw if isinstance(item, dict)]
def _existing_seed_set(existing_entries: list[dict]) -> set[int]:
used: set[int] = set()
for item in existing_entries:
if str(item.get("category") or "").strip() != CATEGORY:
continue
path_text = str(item.get("path") or "").strip()
match = SEED_PATTERN.search(path_text)
if match:
used.add(int(match.group(1)))
video_dir = ROOT / "videos" / CATEGORY
if video_dir.exists():
for file_path in video_dir.glob("*.mp4"):
match = SEED_PATTERN.search(file_path.name)
if match:
used.add(int(match.group(1)))
return used
def _pick_seed_list(count: int, seed_start: int, used: set[int]) -> list[int]:
seeds: list[int] = []
cursor = int(seed_start)
while len(seeds) < int(count):
if cursor not in used:
seeds.append(cursor)
cursor += 1
return seeds
def _upsert_entries(existing_entries: list[dict], new_entries: list[dict]) -> list[dict]:
new_by_path = {
str(entry.get("path") or "").strip(): entry
for entry in new_entries
if str(entry.get("path") or "").strip()
}
out: list[dict] = []
replaced_paths: set[str] = set()
for item in existing_entries:
path_text = str(item.get("path") or "").strip()
if path_text and path_text in new_by_path:
out.append(new_by_path[path_text])
replaced_paths.add(path_text)
else:
out.append(item)
for entry in new_entries:
path_text = str(entry.get("path") or "").strip()
if path_text and path_text not in replaced_paths:
out.append(entry)
return out
def _write_json(path: Path, payload: object) -> None:
path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
def _generate_one(seed: int, force_render: bool = False) -> dict:
out_path = ROOT / "videos" / CATEGORY / f"{CATEGORY}_train_s{int(seed)}.mp4"
out_path.parent.mkdir(parents=True, exist_ok=True)
should_render = bool(force_render) or not (out_path.exists() and out_path.stat().st_size > 0)
rendered_answer = ""
if should_render:
render_cfg = _make_cups_cfg(seed)
_, rendered_answer = generate_cups_shuffle(
str(out_path),
render_cfg,
return_answer=True,
)
trace = _derive_trace(seed)
answer = str(trace["ball_end_under"])
if rendered_answer and rendered_answer != answer:
raise RuntimeError(
f"Answer mismatch for seed={seed}: rendered={rendered_answer}, "
f"trace={answer}"
)
rel_path = out_path.relative_to(ROOT).as_posix()
entry = _build_entry(
rel_path=rel_path,
text=_build_cot_text(trace),
question=_build_question(),
answer=answer,
)
return {
"seed": int(seed),
"status": "rendered" if should_render else "reused",
"entry": entry,
"rel_path": rel_path,
}
def main() -> None:
parser = argparse.ArgumentParser(description="Generate cups_cot train samples.")
parser.add_argument("--count", type=int, default=50, help="Number of new samples to generate.")
parser.add_argument("--seed-start", type=int, default=95000, help="Seed start for this batch.")
parser.add_argument("--workers", type=int, default=0, help="Parallel workers; 0 means auto.")
parser.add_argument(
"--executor",
choices=("thread", "process"),
default="thread",
help="Parallel backend.",
)
parser.add_argument(
"--dataset",
default="video_dataset_train.json",
help="Train dataset JSON path.",
)
parser.add_argument("--force-render", action="store_true", help="Re-render even if target video already exists.")
parser.add_argument("--dry-run", action="store_true", help="Build plan only.")
parser.add_argument(
"--plan-output",
default="tmp_cups_cot_plan.json",
help="Write planned seeds/tasks to this JSON file.",
)
parser.add_argument(
"--records-output",
default="tmp_cups_cot_records.json",
help="Write generation records to this JSON file.",
)
args = parser.parse_args()
if args.count <= 0:
raise SystemExit("--count must be > 0")
dataset_path = Path(args.dataset)
if not dataset_path.is_absolute():
dataset_path = ROOT / dataset_path
plan_path = Path(args.plan_output)
if not plan_path.is_absolute():
plan_path = ROOT / plan_path
records_path = Path(args.records_output)
if not records_path.is_absolute():
records_path = ROOT / records_path
existing_entries = _load_dataset(dataset_path)
used_seeds = _existing_seed_set(existing_entries)
seeds = _pick_seed_list(args.count, args.seed_start, used_seeds)
plan = {
"category": CATEGORY,
"count": int(args.count),
"seed_start_arg": int(args.seed_start),
"seed_list": seeds,
"dataset_path": str(dataset_path),
"executor": args.executor,
}
_write_json(plan_path, plan)
print(f"Existing train entries: {len(existing_entries)}")
print(f"Planned new {CATEGORY} items: {len(seeds)}")
print(f"Seed list: {seeds[0]}..{seeds[-1]}")
print(f"Plan written to: {plan_path}")
if args.dry_run:
return
max_workers = int(args.workers) if int(args.workers) > 0 else min(8, max(1, os.cpu_count() or 1))
print(f"Using workers: {max_workers} ({args.executor})")
executor_cls = ThreadPoolExecutor if args.executor == "thread" else ProcessPoolExecutor
results: list[dict] = []
failures: list[dict] = []
with executor_cls(max_workers=max_workers) as executor:
future_map = {
executor.submit(_generate_one, seed, bool(args.force_render)): seed
for seed in seeds
}
for done_count, future in enumerate(as_completed(future_map), start=1):
seed = future_map[future]
try:
result = future.result()
except Exception as exc: # noqa: BLE001
failures.append({"seed": int(seed), "error": repr(exc)})
else:
results.append(result)
if done_count % 10 == 0 or done_count == len(seeds):
print(f"Progress: {done_count}/{len(seeds)}, failures={len(failures)}")
if failures:
fail_path = ROOT / "tmp_cups_cot_failures.json"
_write_json(fail_path, failures)
raise SystemExit(f"Generation failed for {len(failures)} seeds. See {fail_path}")
results.sort(key=lambda item: int(item["seed"]))
new_entries = [dict(item["entry"]) for item in results]
merged_entries = _upsert_entries(existing_entries, new_entries)
_write_json(dataset_path, merged_entries)
records = {
"category": CATEGORY,
"count": len(results),
"seed_list": [int(item["seed"]) for item in results],
"status_counter": {
"rendered": sum(1 for item in results if item["status"] == "rendered"),
"reused": sum(1 for item in results if item["status"] == "reused"),
},
"records": results,
}
_write_json(records_path, records)
print(f"Wrote dataset: {dataset_path}")
print(f"Wrote records: {records_path}")
print(f"Train size: {len(existing_entries)} -> {len(merged_entries)}")
if __name__ == "__main__":
main()
|