Spaces:
Sleeping
Sleeping
File size: 19,397 Bytes
4ead231 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 | """Pack batch2 (sports) review data.
Differences vs batch1:
- One xlsx (Sport1.xlsx) with FIVE sheets (足球/篮球/排球/艺术体操/游泳).
- Reference images are EMBEDDED inside the xlsx (in xl/media/) and anchored
to specific cells via xl/drawings/drawingN.xml. So we don't read PNGs
from per-video folders for ref_time/ref_space/ref_options; we extract
them from the xlsx.
- Bbox jpg+json still live in the per-video folder under MultiSports/.
- TG_Frames/ is flat: one file <video_id>_<tg_answer>.jpg per anno.
Output:
data/annotations.jsonl one record per row
static/anno_assets/<anno_id>/
ref_col<colidx>_<n>.png embedded refs from xlsx, named by source col
frame_<idx>.jpg bbox keyframes
tg_frames.jpg TG strip from TG_Frames/
Usage:
python scripts/pack_batch2.py \
--batch-root "../batch2/<top-cat-dir>/<inner-cat-dir>" \
--xlsx-name Sport1.xlsx \
--videos-subdir MultiSports \
--tg-frames-subdir TG_Frames \
--out-data data/annotations.jsonl \
--out-assets static/anno_assets
"""
from __future__ import annotations
import argparse
import json
import re
import shutil
import sys
import zipfile
from collections import defaultdict
from pathlib import Path
from typing import Optional
from xml.etree import ElementTree as ET
# Sport sub-category code mapping (sheet name -> filesystem-safe id).
# Names contain CJK chars so we use Unicode escapes in source to be codepage-safe.
SUBCAT_MAP = {
"\u8db3\u7403": "football", # 足球
"\u7bee\u7403": "basketball", # 篮球
"\u6392\u7403": "volleyball", # 排球
"\u827a\u672f\u4f53\u64cd": "artistic_gymnastics", # 艺术体操
"\u6e38\u6cf3": "swimming", # 游泳
}
NS = {
"x": "http://schemas.openxmlformats.org/spreadsheetml/2006/main",
"r": "http://schemas.openxmlformats.org/officeDocument/2006/relationships",
"xdr": "http://schemas.openxmlformats.org/drawingml/2006/spreadsheetDrawing",
"a": "http://schemas.openxmlformats.org/drawingml/2006/main",
"rel": "http://schemas.openxmlformats.org/package/2006/relationships",
}
# ----------------------- xlsx parsing -----------------------
class Xlsx:
"""Cheap-and-cheerful xlsx reader (no openpyxl required)."""
def __init__(self, xlsx_path: Path) -> None:
self.zf = zipfile.ZipFile(xlsx_path)
self._strings: list[str] = []
self._load_shared_strings()
self._workbook = ET.fromstring(self.zf.read("xl/workbook.xml"))
self._wb_rels = self._read_rels("xl/_rels/workbook.xml.rels")
def close(self) -> None:
self.zf.close()
# shared strings ----
def _load_shared_strings(self) -> None:
try:
data = self.zf.read("xl/sharedStrings.xml")
except KeyError:
return
root = ET.fromstring(data)
for si in root.findall("x:si", NS):
ts = si.findall(".//x:t", NS)
self._strings.append("".join((t.text or "") for t in ts))
def s(self, idx: int) -> str:
if 0 <= idx < len(self._strings):
return self._strings[idx]
return ""
# rels ----
def _read_rels(self, path: str) -> dict[str, str]:
try:
data = self.zf.read(path)
except KeyError:
return {}
root = ET.fromstring(data)
out: dict[str, str] = {}
for rel in root.findall("rel:Relationship", NS):
out[rel.get("Id")] = rel.get("Target")
return out
# sheets ----
def sheets(self) -> list[dict]:
"""Return [{name, sheetId, target_path}, ...] in workbook order."""
out = []
for s in self._workbook.findall(".//x:sheets/x:sheet", NS):
rid = s.get(f"{{{NS['r']}}}id")
target = self._wb_rels.get(rid, "")
# Targets are relative to xl/, e.g. "worksheets/sheet1.xml".
target_path = "xl/" + target.lstrip("/")
out.append({
"name": s.get("name", "").strip(),
"sheetId": s.get("sheetId"),
"rid": rid,
"target": target_path,
})
return out
# one sheet ----
def read_sheet(self, sheet_target: str) -> tuple[list[dict], list[dict]]:
"""Return (rows, image_anchors) for a single sheet.
rows is [{ "row": 1-indexed-int, "cells": {col_letter: str}}, ...]
image_anchors is [{"row": 0-idx, "col": 0-idx, "rId": "rId..."}, ...]
"""
sheet_xml = self.zf.read(sheet_target).decode("utf-8")
root = ET.fromstring(sheet_xml)
rows: list[dict] = []
for row in root.findall("x:sheetData/x:row", NS):
rnum = int(row.get("r"))
cells: dict[str, str] = {}
for c in row.findall("x:c", NS):
ref = c.get("r", "")
m = re.match(r"^([A-Z]+)(\d+)$", ref)
if not m:
continue
col = m.group(1)
t = c.get("t", "")
v = c.find("x:v", NS)
if v is None or v.text is None:
continue
if t == "s":
cells[col] = self.s(int(v.text))
else:
cells[col] = v.text
if cells:
rows.append({"row": rnum, "cells": cells})
# ---- drawing ----
# sheet target is like "xl/worksheets/sheet1.xml". Its rels live at
# "xl/worksheets/_rels/sheet1.xml.rels".
sheet_rels_path = sheet_target.replace("worksheets/", "worksheets/_rels/") + ".rels"
sheet_rels = self._read_rels(sheet_rels_path)
anchors: list[dict] = []
for rid, target in sheet_rels.items():
if not target.endswith(".xml") or "drawings/" not in target:
continue
# Resolve drawing path to absolute zip path.
drawing_path = self._resolve_target("xl/worksheets/", target)
drawing_xml = self.zf.read(drawing_path).decode("utf-8")
drawing_root = ET.fromstring(drawing_xml)
drawing_rels_path = drawing_path.replace("drawings/", "drawings/_rels/") + ".rels"
drawing_rels = self._read_rels(drawing_rels_path)
for anc in drawing_root.iter("{%s}oneCellAnchor" % NS["xdr"]):
fr = anc.find("xdr:from", NS)
if fr is None:
continue
col = int(fr.findtext("xdr:col", "0", NS))
row = int(fr.findtext("xdr:row", "0", NS))
blip = anc.find(".//a:blip", NS)
if blip is None:
continue
embed = blip.get(f"{{{NS['r']}}}embed")
image_target = drawing_rels.get(embed, "")
if not image_target:
continue
image_path = self._resolve_target(drawing_path.rsplit("/", 1)[0] + "/", image_target)
anchors.append({
"row": row, "col": col, "image_path": image_path,
})
for anc in drawing_root.iter("{%s}twoCellAnchor" % NS["xdr"]):
fr = anc.find("xdr:from", NS)
if fr is None:
continue
col = int(fr.findtext("xdr:col", "0", NS))
row = int(fr.findtext("xdr:row", "0", NS))
blip = anc.find(".//a:blip", NS)
if blip is None:
continue
embed = blip.get(f"{{{NS['r']}}}embed")
image_target = drawing_rels.get(embed, "")
if not image_target:
continue
image_path = self._resolve_target(drawing_path.rsplit("/", 1)[0] + "/", image_target)
anchors.append({
"row": row, "col": col, "image_path": image_path,
})
return rows, anchors
def _resolve_target(self, base_dir: str, target: str) -> str:
"""Resolve a relationship target relative to the file declaring it."""
# Targets often start with "../"; collapse against base_dir.
parts = (base_dir + target).split("/")
out: list[str] = []
for p in parts:
if p == "..":
if out:
out.pop()
elif p in ("", "."):
continue
else:
out.append(p)
return "/".join(out)
def extract_image(self, archive_path: str, dst_path: Path) -> None:
dst_path.parent.mkdir(parents=True, exist_ok=True)
with self.zf.open(archive_path) as src, dst_path.open("wb") as dst:
shutil.copyfileobj(src, dst)
# ----------------------- bbox + tg frames -----------------------
def load_bboxes(video_dir: Path) -> list[dict]:
"""Read all *.json next to *.jpg in this dir; return parsed bbox metadata."""
out = []
for js in sorted(video_dir.glob("*.json")):
try:
doc = json.loads(js.read_text(encoding="utf-8"))
except Exception:
continue
m = re.search(r"_(\d+)\.json$", js.name)
frame_idx = int(m.group(1)) if m else None
boxes = []
for s in doc.get("shapes", []):
if s.get("shape_type") != "rectangle":
continue
pts = s.get("points") or []
if len(pts) < 2:
continue
(x1, y1), (x2, y2) = pts[0], pts[1]
if x1 > x2: x1, x2 = x2, x1
if y1 > y2: y1, y2 = y2, y1
boxes.append({
"label": str(s.get("label", "?")),
"x1": round(x1, 2), "y1": round(y1, 2),
"x2": round(x2, 2), "y2": round(y2, 2),
})
# find matching jpg
jpg = video_dir / (doc.get("imagePath") or js.with_suffix(".jpg").name)
out.append({
"frame_idx": frame_idx,
"jpg_path": jpg if jpg.exists() else None,
"image_w": doc.get("imageWidth"),
"image_h": doc.get("imageHeight"),
"boxes": boxes,
})
out.sort(key=lambda x: x["frame_idx"] if x["frame_idx"] is not None else 0)
return out
def index_tg_frames(tg_root: Path) -> dict[str, Path]:
"""Build {video_id -> jpg_path}. Filenames are <video_id>_<answer>.jpg."""
if not tg_root.exists():
return {}
items = list(tg_root.rglob("*.jpg"))
return {p.stem: p for p in items} # we'll do prefix-match later via lookup
def resolve_tg_frame(tg_index: dict[str, Path], video_id: str) -> Optional[Path]:
"""Find the TG frame strip whose stem starts with '<video_id>_'."""
needle = video_id + "_"
matches = [p for stem, p in tg_index.items() if stem.startswith(needle)]
if not matches:
return None
return min(matches, key=lambda p: len(p.stem))
# ----------------------- main -----------------------
def main() -> None:
ap = argparse.ArgumentParser()
ap.add_argument("--batch-root", required=True,
help="dir containing the xlsx + MultiSports/ + TG_Frames/")
ap.add_argument("--xlsx-name", default="Sport1.xlsx")
ap.add_argument("--videos-subdir", default="MultiSports")
ap.add_argument("--tg-frames-subdir", default="TG_Frames")
ap.add_argument("--out-data", required=True)
ap.add_argument("--out-assets", required=True)
args = ap.parse_args()
batch_root = Path(args.batch_root).resolve()
xlsx_path = batch_root / args.xlsx_name
videos_root = batch_root / args.videos_subdir
tg_root = batch_root / args.tg_frames_subdir
if not xlsx_path.exists():
print(f"[fatal] xlsx not found: {xlsx_path}")
sys.exit(1)
out_data = Path(args.out_data).resolve()
out_assets = Path(args.out_assets).resolve()
out_data.parent.mkdir(parents=True, exist_ok=True)
out_assets.mkdir(parents=True, exist_ok=True)
tg_index = index_tg_frames(tg_root)
print(f"[tg] indexed {len(tg_index)} TG strips from {tg_root}")
print(f"[xlsx] opening {xlsx_path}")
xlsx = Xlsx(xlsx_path)
sheets = xlsx.sheets()
print(f"[xlsx] {len(sheets)} sheets:")
for s in sheets:
print(f" - {s['name']!r} (sheetId={s['sheetId']}, target={s['target']})")
annotations: list[dict] = []
skipped: list[dict] = []
for sh in sheets:
sub_id = SUBCAT_MAP.get(sh["name"])
if sub_id is None:
print(f"[warn] unknown sheet name {sh['name']!r}, skipping")
continue
rows, anchors = xlsx.read_sheet(sh["target"])
# Group anchors by row index (0-based) -> list of (col, image_path)
anchors_by_row: dict[int, list[dict]] = defaultdict(list)
for a in anchors:
anchors_by_row[a["row"]].append(a)
# Sort each row by col so smaller col index comes first (matches B/D/F order).
for r in anchors_by_row:
anchors_by_row[r].sort(key=lambda a: a["col"])
print(f"[sheet:{sub_id}] {len(rows)} rows, {len(anchors)} image anchors")
for r in rows:
if r["row"] == 1:
continue # header
cells = r["cells"]
raw_video_id = (cells.get("A") or "").strip()
if not raw_video_id:
continue
# Skip group separator rows like cell A == "1", "2", "3" etc.
if re.fullmatch(r"\d+", raw_video_id):
continue
# Skip rows where col A is a CJK label (e.g. "沙滩排球" subcategory header).
if not re.search(r"[A-Za-z0-9_]", raw_video_id):
continue
# Normalize the id to find the on-disk folder:
# - Excel sometimes wraps a leading-dash id in full-width parens,
# e.g. "-edKF31R_Tk" -> "(-)edKF31R_Tk"
# - Some labelers leave a stray trailing "=".
video_id = raw_video_id.rstrip("=")
video_id = video_id.replace("\uFF08\u002D\uFF09", "-") # (-) -> -
video_id = video_id.replace("\uFF08\uFF0D\uFF09", "-") # (-) -> -
anno_id = f"sport__{sub_id}__{video_id}"
asset_dir = out_assets / anno_id
asset_dir.mkdir(parents=True, exist_ok=True)
# ---- embedded images from xlsx, indexed by source column ----
# In the sheet, xdr:row is 0-based, so the row that contains
# 1-indexed row 2 has xdr:row==1. Anchors usually attach to the
# row's TOP edge.
row_anchors = anchors_by_row.get(r["row"] - 1, [])
ref_files: list[dict] = []
# Group by which "slot" they look like:
# slot determined by xdr:col index (B=1, D=3, F=5 in sheet order).
COL_SLOT = {1: "ref_time", 3: "ref_space", 5: "ref_options"}
saved: dict[str, str] = {}
for i, a in enumerate(row_anchors):
slot = COL_SLOT.get(a["col"])
if slot and slot not in saved:
dst = asset_dir / f"{slot}.png"
xlsx.extract_image(a["image_path"], dst)
saved[slot] = dst.name
else:
# extra image; preserve under a generic name
dst = asset_dir / f"ref_extra_col{a['col']}_{i}.png"
xlsx.extract_image(a["image_path"], dst)
# ---- bbox frames from MultiSports/<video_id>/ ----
video_dir = videos_root / video_id
frames_meta: list[dict] = []
if video_dir.exists():
bboxes = load_bboxes(video_dir)
for bb in bboxes:
idx = bb["frame_idx"] if bb["frame_idx"] is not None else 0
dst_jpg = asset_dir / f"frame_{idx}.jpg"
if bb["jpg_path"]:
shutil.copyfile(bb["jpg_path"], dst_jpg)
frames_meta.append({
"frame_idx": bb["frame_idx"],
"image": dst_jpg.name,
"image_w": bb["image_w"],
"image_h": bb["image_h"],
"boxes": bb["boxes"],
})
else:
skipped.append({"anno_id": anno_id, "reason": "video dir missing", "video_id": video_id})
# ---- TG strip ----
tg_jpg = resolve_tg_frame(tg_index, video_id)
tg_name = None
if tg_jpg is not None:
dst = asset_dir / "tg_frames.jpg"
shutil.copyfile(tg_jpg, dst)
tg_name = dst.name
# ---- options blob -> dict ----
options_blob = (cells.get("G") or "")
qa_options: dict[str, str] = {}
for line in re.split(r"[\r\n]+", options_blob):
line = line.strip()
if not line:
continue
m = re.match(r"^([A-EA-E])[\s\.\:、]?\s*(.+)$", line)
if m:
key = m.group(1)
if "\uFF21" <= key <= "\uFF25":
key = chr(ord(key) - 0xFF21 + ord("A"))
qa_options[key] = m.group(2).strip()
rec = {
"anno_id": anno_id,
"category": "sports",
"subcategory": sub_id,
"subcategory_zh": sh["name"],
"video_id": video_id,
"tg_question": cells.get("B") or "",
"tg_answer": cells.get("C") or "",
"sg_question": cells.get("D") or "",
"sg_answer": cells.get("E") or "", # batch2 actually fills this
"qa_question": cells.get("F") or "",
"qa_options": qa_options,
"qa_answer": cells.get("H") or "",
"extra_col_i": cells.get("I") or "", # diagnostic
"ref_time": saved.get("ref_time"),
"ref_space": saved.get("ref_space"),
"ref_options": saved.get("ref_options"),
"tg_frames": tg_name,
"frames": frames_meta,
"xlsx_row": r["row"],
"xlsx_source": "Sport1.xlsx",
}
annotations.append(rec)
xlsx.close()
# ---- write jsonl, ASCII-safe ----
out_data.write_text(
"\n".join(json.dumps(a, ensure_ascii=True) for a in annotations) + "\n",
encoding="utf-8",
)
# ---- stats ----
print()
print(f"TOTAL annotations: {len(annotations)}")
print(f" skipped (video dir missing): {len(skipped)}")
by_sub: dict[str, int] = defaultdict(int)
for a in annotations:
by_sub[a["subcategory"]] += 1
for k, n in sorted(by_sub.items(), key=lambda kv: -kv[1]):
print(f" {k}: {n}")
print()
def count_field(field: str) -> int:
return sum(1 for a in annotations if a.get(field))
for f in ("ref_time", "ref_space", "ref_options", "tg_frames", "sg_answer"):
print(f" {f}: {count_field(f)} / {len(annotations)}")
print()
print(f"wrote {out_data}")
print(f"wrote {out_assets}/ ({len(annotations)} anno dirs)")
if skipped:
print()
print("=== skipped (first 10) ===")
for s in skipped[:10]:
print(f" {s['anno_id']}: {s['reason']}")
if __name__ == "__main__":
main()
|