import os, json, argparse, logging, csv from typing import List, Dict, Any import pandas as pd from datasets import Dataset, DatasetDict, Features, Value, Image, Sequence from huggingface_hub import login, HfApi logging.basicConfig(level=logging.INFO) logger = logging.getLogger("viinfographicvqa") # ----------------- helpers ----------------- def read_json_list(path: str) -> List[Dict[str, Any]]: with open(path, "r", encoding="utf-8") as f: data = json.load(f) assert isinstance(data, list), f"{path} phải là list các object" return data def ensure_list(x): if x is None: return [] return x if isinstance(x, list) else [x] def exists(p: str) -> bool: return bool(p) and os.path.exists(p) def write_missing(rows: List[Dict[str, Any]], out_path: str): if not rows: return keys = rows[0].keys() with open(out_path, "w", newline="", encoding="utf-8") as f: w = csv.DictWriter(f, fieldnames=keys); w.writeheader(); w.writerows(rows) logger.warning(f"⚠️ Báo cáo ảnh thiếu: {out_path} ({len(rows)} dòng)") # ----------------- builders ----------------- def build_single_df(items: List[Dict[str, Any]]) -> pd.DataFrame: rows = [] for ex in items: img = ex.get("image_path") basename = os.path.basename(img) if img else None rows.append({ "question_id" : str(ex.get("question_id")), "images_paths" : [basename] if basename else [], # chỉ tên file "image_type" : ex.get("image_type"), "answer_source": ex.get("answer_source"), "element" : ensure_list(ex.get("element")), "question" : ex.get("question"), "answer" : ex.get("answer"), }) return pd.DataFrame(rows) def build_multi_df(items: List[Dict[str, Any]]) -> pd.DataFrame: rows = [] for ex in items: basenames = [os.path.basename(p) for p in ex.get("image_paths", [])] rows.append({ "question_id" : str(ex.get("question_id")), "images_paths" : basenames, # chỉ tên file "image_type" : ex.get("image_type"), "answer_source": ex.get("answer_source"), "element" : ensure_list(ex.get("element")) if "element" in ex else [], "question" : ex.get("question"), "answer" : ex.get("answer"), }) return pd.DataFrame(rows) def make_unified_dataset(df: pd.DataFrame, base_dir: str, split_name: str, images_dirname: str="images") -> Dataset: """images_paths = list[str] (filenames only). preview image read from images/{filename}.""" df = df.copy() # preview path: lấy ảnh đầu tiên df["preview_path"] = df["images_paths"].map(lambda lst: os.path.join(base_dir, images_dirname, lst[0]) if (lst and lst[0]) else None) # validate: mọi filename phải tồn tại trong images/ missing = [] keep = [] for i, filenames in enumerate(df["images_paths"]): # full paths fulls = [os.path.join(base_dir, images_dirname, fn) for fn in (filenames or [])] ok = bool(fulls) and all(exists(p) for p in fulls) if not ok: miss = [p for p in fulls if not exists(p)] missing.append({ "split": split_name, "row_index": i, "question_id": df.loc[i, "question_id"], "missing": ";".join(miss) if miss else "(no images)" }) keep.append(ok) if any(not k for k in keep): logger.warning(f"[{split_name}] Bỏ {sum(1 for k in keep if not k)} mẫu thiếu ảnh") df = df[keep].reset_index(drop=True) # unified features cho tất cả split features = Features({ "question_id" : Value("string"), "images_paths" : Sequence(Value("string")), # chỉ tên file "image" : Image(), # preview (nhúng bytes vào parquet) "image_type" : Value("string"), "answer_source": Value("string"), "element" : Sequence(Value("string")), "question" : Value("string"), "answer" : Value("string"), }) # tạo dataset: rename preview_path -> image rồi cast ds = Dataset.from_pandas( df[["question_id","images_paths","preview_path","image_type","answer_source","element","question","answer"]], preserve_index=False ) ds = ds.rename_column("preview_path", "image").cast(features) return ds, missing # ----------------- uploader ----------------- def upload_images_folder(repo_id: str, images_dir: str, path_in_repo: str="images", repo_type: str="dataset"): """Upload toàn bộ thư mục images/ lên repo dataset. LƯU Ý: nhiều file nhỏ => thời gian lâu. Đã khuyến nghị bật HF_TRANSFER. """ logger.info(f"Uploading folder '{images_dir}' to '{repo_id}/{path_in_repo}' ...") api = HfApi() api.upload_folder( folder_path=images_dir, repo_id=repo_id, repo_type=repo_type, path_in_repo=path_in_repo, commit_message="Upload raw images folder", allow_patterns=None, # hoặc ví dụ ["*.jpg","*.png"] ignore_patterns=None, ) logger.info("✅ Upload images folder done.") # ----------------- main ----------------- def main(): ap = argparse.ArgumentParser(description="Push ViInfographicVQA (unified schema, filenames only) to HF Hub") ap.add_argument("--repo_id", required=True, help="VD: VLAI-AIVN/ViInfographicVQA") ap.add_argument("--hf_token", default=None) ap.add_argument("--base_dir", default=".", help="Thư mục chứa /images và /data") ap.add_argument("--images_subdir", default="images", help="Tên thư mục ảnh (mặc định: images)") ap.add_argument("--max_shard_size", default="4GB") ap.add_argument("--private", action="store_true") ap.add_argument("--branch", default=None) # ví dụ: parquet-v1 ap.add_argument("--dry_run", action="store_true") ap.add_argument("--upload_images_folder", action="store_true", help="Sau khi push parquet, upload cả thư mục images/ lên repo") args = ap.parse_args() if args.hf_token: login(token=args.hf_token) base = os.path.abspath(args.base_dir) data_dir = os.path.join(base, "data") images_dir = os.path.join(base, args.images_subdir) if not os.path.isdir(images_dir): raise FileNotFoundError(f"Không thấy {args.images_subdir}/: {images_dir}") if not os.path.isdir(data_dir): raise FileNotFoundError(f"Không thấy data/: {data_dir}") logger.info(f"Base: {base}") logger.info(f"Images: {len(os.listdir(images_dir))} files") # đọc annotations def read_if(path): return read_json_list(path) if os.path.exists(path) else None st = read_if(os.path.join(data_dir, "single_train.json")) sv = read_if(os.path.join(data_dir, "single_test.json")) mt = read_if(os.path.join(data_dir, "multi_train.json")) mv = read_if(os.path.join(data_dir, "multi_test.json")) ddict = {} missing_all = [] if st: from_df, miss = make_unified_dataset(build_single_df(st), base, "single_train", args.images_subdir) ddict["single_train"] = from_df; missing_all += miss logger.info(f"single_train: {len(from_df)} samples") if sv: from_df, miss = make_unified_dataset(build_single_df(sv), base, "single_test", args.images_subdir) ddict["single_test"] = from_df; missing_all += miss logger.info(f"single_test : {len(from_df)} samples") if mt: from_df, miss = make_unified_dataset(build_multi_df(mt), base, "multi_train", args.images_subdir) ddict["multi_train"] = from_df; missing_all += miss logger.info(f"multi_train : {len(from_df)} samples") if mv: from_df, miss = make_unified_dataset(build_multi_df(mv), base, "multi_test", args.images_subdir) ddict["multi_test"] = from_df; missing_all += miss logger.info(f"multi_test : {len(from_df)} samples") if not ddict: raise SystemExit("Không có split nào để push (thiếu file json?).") # báo cáo ảnh thiếu write_missing(missing_all, os.path.join(base, "missing_images_report.csv")) if args.dry_run: logger.info("Dry run OK, không push parquet / images."); return # push 4 split cùng lúc (đảm bảo schema đồng nhất) push_kwargs = dict( max_shard_size=args.max_shard_size, private=args.private, commit_message="Initial upload: unified schema (images_paths=filenames, preview image) → Parquet shards", ) if args.branch: push_kwargs["revision"] = args.branch ds_all = DatasetDict(ddict) logger.info(f"Pushing parquet shards to {args.repo_id} ...") ds_all.push_to_hub(args.repo_id, **push_kwargs) logger.info("✅ Parquet splits uploaded.") # (khuyến nghị) upload thư mục images/ để người dùng tải đủ ảnh if args.upload_images_folder: upload_images_folder(args.repo_id, images_dir, path_in_repo=args.images_subdir, repo_type="dataset") logger.info("🎉 Hoàn tất.") if __name__ == "__main__": main()