wkea's picture
Deploy Space
c388140 verified
Raw History Blame Contribute Delete
2.33 kB
from __future__ import annotations
import csv
import json
import os
from dataclasses import dataclass
from typing import Iterable, List, Sequence, Tuple
@dataclass(frozen=True)
class CaptionItem:
"""
一条图文配对样本:
- image_path:图片路径(绝对路径)
- text:文本描述/提示词
"""
image_path: str
text: str
def load_captions(captions_path: str, data_root: str) -> List[CaptionItem]:
"""
读取 caption 文件,支持:
- .jsonl:每行 {"file": "...", "text": "..."},file 可相对 data_root 或绝对路径
- .csv:两列 file,text(含表头),file 同上
"""
captions_path = os.path.abspath(captions_path)
data_root = os.path.abspath(data_root)
if not os.path.exists(captions_path):
raise FileNotFoundError(f"未找到 captions 文件:{captions_path}")
items: List[CaptionItem] = []
ext = os.path.splitext(captions_path)[1].lower()
def resolve_path(p: str) -> str:
p = (p or "").strip()
if not p:
return ""
if os.path.isabs(p):
return p
return os.path.join(data_root, p)
if ext == ".jsonl":
with open(captions_path, "r", encoding="utf-8") as f:
for line in f:
line = line.strip()
if not line:
continue
obj = json.loads(line)
img = resolve_path(str(obj.get("file", "")))
txt = str(obj.get("text", ""))
if img and os.path.exists(img):
items.append(CaptionItem(image_path=img, text=txt))
elif ext == ".csv":
with open(captions_path, "r", encoding="utf-8") as f:
reader = csv.DictReader(f)
for row in reader:
img = resolve_path(str(row.get("file", "")))
txt = str(row.get("text", ""))
if img and os.path.exists(img):
items.append(CaptionItem(image_path=img, text=txt))
else:
raise ValueError("captions 文件仅支持 .jsonl 或 .csv")
if len(items) == 0:
raise ValueError("captions 文件未读取到有效样本(请检查 file 路径与图片是否存在)")
return items