File size: 12,439 Bytes
f9ed2b4 | 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 | #!/usr/bin/env python3
'Build seeded binary and source-classification training subsets from fixed image pools.'
from __future__ import annotations
import argparse
import csv
import hashlib
import json
import sys
from collections import defaultdict
from pathlib import Path
from typing import Any
PROJECT_ROOT = Path(__file__).resolve().parents[2]
DEFAULT_SPLIT_DIR = PROJECT_ROOT / "artifacts" / "metadata" / "fixed_split"
DEFAULT_OUTPUT_DIR = PROJECT_ROOT / "artifacts" / "metadata" / "training"
DEFAULT_N_GRID = [900, 90, 45, 18]
csv.field_size_limit(min(sys.maxsize, 2_147_483_647))
def stable_rank(seed: int, group_key: str, sample_id: str) -> int:
key = f"{seed}:{group_key}:{sample_id}".encode("utf-8")
return int(hashlib.sha256(key).hexdigest(), 16)
def read_csv(path: Path) -> list[dict[str, str]]:
with path.open("r", newline="", encoding="utf-8") as f:
return list(csv.DictReader(f))
def write_json(path: Path, data: Any) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
with path.open("w", encoding="utf-8") as f:
json.dump(data, f, indent=2, ensure_ascii=False)
def parse_n_grid(value: str) -> list[int]:
n_grid = [int(part.strip()) for part in value.split(",") if part.strip()]
if not n_grid:
raise ValueError("N grid cannot be empty")
return sorted(n_grid, reverse=True)
def group_ai(rows: list[dict[str, str]]) -> dict[str, list[dict[str, str]]]:
grouped: dict[str, list[dict[str, str]]] = defaultdict(list)
for row in rows:
grouped[row["source_model_id"]].append(row)
return grouped
def group_real_sources(rows: list[dict[str, str]]) -> dict[str, list[dict[str, str]]]:
grouped: dict[str, list[dict[str, str]]] = defaultdict(list)
for row in rows:
grouped[row["real_source_id"]].append(row)
return grouped
def ranked_rows(rows: list[dict[str, str]], seed: int, group: str) -> list[dict[str, str]]:
"""Independent seeded training ranks; validation order is seed-independent."""
if "train" in group:
return sorted(rows, key=lambda row: (stable_rank(seed, group, row["sample_id"]), row["sample_id"]))
return sorted(rows, key=lambda row: (int(row.get("split_rank") or 0), row["sample_id"]))
def select_real_rows(
rows: list[dict[str, str]],
total: int,
seed: int,
) -> list[dict[str, str]]:
ranked = sorted(
rows,
key=lambda row: (
stable_rank(seed, f"real_train:{row.get('source_dataset', 'unknown')}", row["sample_id"]),
row["sample_id"],
),
)
if total > len(ranked):
raise RuntimeError(f"Requested {total} real rows, but only {len(ranked)} are available")
return ranked[:total]
def real_source_counts(total: int, source_ids: list[str]) -> dict[str, int]:
base_count = total // len(source_ids)
remainder = total % len(source_ids)
return {
source_id: base_count + (1 if i < remainder else 0)
for i, source_id in enumerate(source_ids)
}
def to_path_list(rows: list[dict[str, str]]) -> list[str]:
return [row["image_path"] for row in rows]
def build_split(
ai_train_rows: list[dict[str, str]],
ai_val_rows: list[dict[str, str]],
real_train_rows: list[dict[str, str]],
real_val_rows: list[dict[str, str]],
n_grid: list[int],
real_ratio: float,
seed: int,
) -> tuple[dict[str, dict[str, list[str]]], dict[str, Any]]:
max_n = max(n_grid)
ai_train_by_model = group_ai(ai_train_rows)
ai_val_by_model = group_ai(ai_val_rows)
train: dict[str, list[str]] = {}
test: dict[str, list[str]] = {}
model_summaries: list[dict[str, Any]] = []
for model_id in sorted(ai_train_by_model, key=lambda x: int(x)):
ranked_train = ranked_rows(ai_train_by_model[model_id], seed, f"ai_train:{model_id}")
ranked_val = ranked_rows(ai_val_by_model.get(model_id, []), seed, f"ai_val:{model_id}")
if len(ranked_train) < max_n:
raise RuntimeError(f"Model {model_id} has {len(ranked_train)} train rows, expected at least {max_n}")
train[model_id] = to_path_list(ranked_train[:max_n])
test[model_id] = to_path_list(ranked_val)
model_summaries.append(
{
"source_model_id": int(model_id),
"train_images": len(train[model_id]),
"validation_images": len(test[model_id]),
}
)
max_real_train = int(real_ratio * max_n * len(model_summaries))
selected_real_train = select_real_rows(real_train_rows, max_real_train, seed)
train["0"] = to_path_list(selected_real_train)
test["0"] = to_path_list(real_val_rows)
split = {"train": train, "test": test, "protocol": {"seed": seed, "version": "release-v1-resampled", "checkpoint": "final"}}
summary = {
"seed": seed,
"n_grid": n_grid,
"max_n": max_n,
"real_ratio": real_ratio,
"ai_model_count": len(model_summaries),
"ai_train_images_at_max_n": max_n * len(model_summaries),
"real_train_images_at_max_n": max_real_train,
"ai_validation_images": sum(len(v) for k, v in test.items() if k != "0"),
"real_validation_images": len(test["0"]),
"models": model_summaries,
"per_n": [
{
"n": n,
"ai_train_images": n * len(model_summaries),
"real_train_images": int(real_ratio * n * len(model_summaries)),
"total_train_images": n * len(model_summaries) + int(real_ratio * n * len(model_summaries)),
}
for n in n_grid
],
}
return split, summary
def build_source_multiclass_split(
ai_train_rows: list[dict[str, str]],
ai_val_rows: list[dict[str, str]],
real_train_rows: list[dict[str, str]],
real_val_rows: list[dict[str, str]],
n_grid: list[int],
real_ratio: float,
seed: int,
) -> tuple[dict[str, dict[str, list[str]]], dict[str, Any]]:
max_n = max(n_grid)
ai_train_by_model = group_ai(ai_train_rows)
ai_val_by_model = group_ai(ai_val_rows)
real_train_by_source = group_real_sources(real_train_rows)
real_val_by_source = group_real_sources(real_val_rows)
train: dict[str, list[str]] = {}
test: dict[str, list[str]] = {}
model_summaries: list[dict[str, Any]] = []
for model_id in sorted(ai_train_by_model, key=lambda x: int(x)):
ranked_train = ranked_rows(ai_train_by_model[model_id], seed, f"ai_train:{model_id}")
ranked_val = ranked_rows(ai_val_by_model.get(model_id, []), seed, f"ai_val:{model_id}")
if len(ranked_train) < max_n:
raise RuntimeError(f"Model {model_id} has {len(ranked_train)} train rows, expected at least {max_n}")
train[model_id] = to_path_list(ranked_train[:max_n])
test[model_id] = to_path_list(ranked_val)
model_summaries.append(
{
"source_model_id": int(model_id),
"train_images": len(train[model_id]),
"validation_images": len(test[model_id]),
}
)
real_source_ids = sorted(real_train_by_source, key=lambda x: int(x))
max_real_train = int(real_ratio * max_n * len(model_summaries))
per_source_max_counts = real_source_counts(max_real_train, real_source_ids)
real_source_summaries: list[dict[str, Any]] = []
for source_id in real_source_ids:
ranked_train = ranked_rows(
real_train_by_source[source_id],
seed,
f"real_train_source:{source_id}",
)
limit = per_source_max_counts[source_id]
if len(ranked_train) < limit:
raise RuntimeError(
f"Real source {source_id} has {len(ranked_train)} train rows, expected at least {limit}"
)
train[source_id] = to_path_list(ranked_train[:limit])
test[source_id] = to_path_list(
ranked_rows(real_val_by_source.get(source_id, []), seed, f"real_val_source:{source_id}")
)
real_source_summaries.append(
{
"real_source_id": int(source_id),
"train_images": len(train[source_id]),
"validation_images": len(test[source_id]),
}
)
split = {"train": train, "test": test, "protocol": {"seed": seed, "version": "release-v1-resampled", "checkpoint": "final"}}
summary = {
"seed": seed,
"n_grid": n_grid,
"max_n": max_n,
"real_ratio": real_ratio,
"ai_model_count": len(model_summaries),
"real_source_count": len(real_source_summaries),
"source_class_count": len(model_summaries) + len(real_source_summaries),
"ai_train_images_at_max_n": max_n * len(model_summaries),
"real_train_images_at_max_n": max_real_train,
"ai_validation_images": sum(len(v) for k, v in test.items() if int(k) > 0),
"real_validation_images": sum(len(v) for k, v in test.items() if int(k) < 0),
"models": model_summaries,
"real_sources": real_source_summaries,
"per_n": [
{
"n": n,
"ai_train_images": n * len(model_summaries),
"real_train_images": int(real_ratio * n * len(model_summaries)),
"total_train_images": n * len(model_summaries) + int(real_ratio * n * len(model_summaries)),
}
for n in n_grid
],
}
return split, summary
def main() -> None:
parser = argparse.ArgumentParser(description="Build WildGen training split configs.")
parser.add_argument("--ai-train", type=Path, default=DEFAULT_SPLIT_DIR / "ai_train_pool.csv")
parser.add_argument("--ai-val", type=Path, default=DEFAULT_SPLIT_DIR / "ai_validation.csv")
parser.add_argument("--real-train", type=Path, default=DEFAULT_SPLIT_DIR / "real_train_pool.csv")
parser.add_argument("--real-val", type=Path, default=DEFAULT_SPLIT_DIR / "real_validation.csv")
parser.add_argument("--output-dir", type=Path, default=DEFAULT_OUTPUT_DIR / "splits")
parser.add_argument("--n-grid", type=str, default=",".join(str(n) for n in DEFAULT_N_GRID))
parser.add_argument("--real-ratio", type=float, default=0.1)
parser.add_argument("--seed", type=int, default=42)
args = parser.parse_args()
n_grid = parse_n_grid(args.n_grid)
ai_train_rows = read_csv(args.ai_train)
ai_val_rows = read_csv(args.ai_val)
real_train_rows = read_csv(args.real_train)
real_val_rows = read_csv(args.real_val)
split, summary = build_split(
ai_train_rows,
ai_val_rows,
real_train_rows,
real_val_rows,
n_grid=n_grid,
real_ratio=args.real_ratio,
seed=args.seed,
)
source_split, source_summary = build_source_multiclass_split(
ai_train_rows,
ai_val_rows,
real_train_rows,
real_val_rows,
n_grid=n_grid,
real_ratio=args.real_ratio,
seed=args.seed,
)
ratio_tag = str(args.real_ratio).replace(".", "p")
split_path = args.output_dir / f"training_split_n{max(n_grid)}_realratio_{ratio_tag}_seed{args.seed}.json"
summary_path = args.output_dir / f"training_split_n{max(n_grid)}_realratio_{ratio_tag}_seed{args.seed}_summary.json"
source_split_path = args.output_dir / f"source_split_n{max(n_grid)}_realratio_{ratio_tag}_seed{args.seed}.json"
source_summary_path = args.output_dir / f"source_split_n{max(n_grid)}_realratio_{ratio_tag}_seed{args.seed}_summary.json"
write_json(split_path, split)
summary["split_file"] = str(split_path)
write_json(summary_path, summary)
write_json(source_split_path, source_split)
source_summary["split_file"] = str(source_split_path)
write_json(source_summary_path, source_summary)
print(json.dumps({
"binary_split_file": str(split_path),
"binary_summary_file": str(summary_path),
"source_split_file": str(source_split_path),
"source_summary_file": str(source_summary_path),
"binary": summary,
"source_multiclass": source_summary,
}, indent=2))
if __name__ == "__main__":
main()
|