Datasets:
File size: 6,107 Bytes
53de216 | 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 | """Source adapters: read each source's native annotation format into one row per image.
Every adapter returns a DataFrame with columns
rel_path path relative to <raw_root>/<source_id>
native_label native tokens joined by '|' (kept verbatim next to the harmonized label)
patient_id patient / exam identifier when the source provides one, else ''
eye left / right / ''
dr_grade ICDR grade 0-4 as string when available, else ''
"""
import re
from pathlib import Path
import pandas as pd
IMG_EXT = {".jpg", ".jpeg", ".png", ".tif", ".tiff", ".bmp"}
COLS = ["rel_path", "native_label", "patient_id", "eye", "dr_grade"]
def _norm(v):
if v is None or (isinstance(v, float) and pd.isna(v)):
return ""
s = str(v).strip()
if s.lower() == "nan":
return ""
if re.fullmatch(r"-?\d+\.0+", s):
s = s.split(".")[0]
return s
def _first(row, cols):
for c in [cols] if isinstance(cols, str) else cols:
v = _norm(row.get(c))
if v:
return v
return ""
def _need(df, cols, where):
cols = [c for c in cols if c]
missing = [c for c in cols if c not in df.columns]
if missing:
raise KeyError(f"{where}: missing columns {missing}; found {list(df.columns)[:30]}")
def _read_table(path, sep=",", header=True, names=None):
path = Path(path)
if path.suffix.lower() in {".xlsx", ".xls"}:
df = pd.read_excel(path, dtype=str)
else:
df = pd.read_csv(path, sep=sep, header=0 if header else None, names=names, dtype=str,
skipinitialspace=True, engine="python")
df.columns = [str(c).strip() for c in df.columns]
return df
def _flatten(x):
if not x:
return []
return [x] if isinstance(x, str) else list(x)
def adapter_csv(src_dir, cfg):
rows = []
for t in cfg["tables"]:
df = _read_table(src_dir / t["file"], t.get("sep", ","), t.get("header", True), t.get("names"))
label_cols = _flatten(cfg.get("label_col"))
mh = cfg.get("multihot_cols")
if mh is None and cfg.get("multihot_exclude_cols") is not None:
mh = [c for c in df.columns if c not in cfg["multihot_exclude_cols"]]
_need(df, [cfg["image_col"], cfg.get("patient_col"), cfg.get("eye_col"), *label_cols,
*_flatten(cfg.get("grade_col")), *(mh or []), *(cfg.get("keep_if") or {})], t["file"])
for _, r in df.iterrows():
if any(_norm(r[c]) != _norm(v) for c, v in (cfg.get("keep_if") or {}).items()):
continue
img = _norm(r[cfg["image_col"]])
if not img:
continue
if label_cols:
tokens = [_first(r, label_cols)]
else:
tokens = [c for c in mh if _norm(r[c]) == "1"]
if not tokens and cfg.get("normal_if_all_zero"):
tokens = ["NORMAL"]
patient = _norm(r[cfg["patient_col"]]) if cfg.get("patient_col") else ""
if not patient and cfg.get("patient_regex"):
m = re.search(cfg["patient_regex"], img)
patient = m.group(1) if m else ""
eye = _norm(r[cfg["eye_col"]]) if cfg.get("eye_col") else ""
if not eye and cfg.get("eye_regex"):
m = re.search(cfg["eye_regex"], img)
eye = m.group(1) if m else ""
rel = Path(t["image_dir"]) / cfg.get("path_template", "{image}").format(image=img, patient=patient)
grade = _first(r, cfg["grade_col"]) if cfg.get("grade_col") else ""
rows.append([str(rel), "|".join(tokens), patient, eye.lower(), grade])
return pd.DataFrame(rows, columns=COLS)
def _images_under(base, pattern="**/*"):
return sorted(p for p in base.glob(pattern) if p.is_file() and p.suffix.lower() in IMG_EXT)
def adapter_folders(src_dir, cfg):
"""Class = name of the image's parent folder (e.g. train/glaucoma/x.png -> 'glaucoma')."""
base = src_dir / cfg.get("subdir", "")
rows = [[str(p.relative_to(src_dir)), p.parent.name, "", "", ""] for p in _images_under(base)]
return pd.DataFrame(rows, columns=COLS)
def adapter_filename(src_dir, cfg):
"""Class from regex rules on the file name; first matching rule wins, unmatched files get no label."""
rules = [(re.compile(p), tok) for p, tok in cfg["rules"]]
rows = []
for p in _images_under(src_dir, cfg.get("glob", "**/*")):
tok = next((tok for rx, tok in rules if rx.search(p.name)), "")
rows.append([str(p.relative_to(src_dir)), tok, "", "", ""])
return pd.DataFrame(rows, columns=COLS)
def adapter_odir(src_dir, cfg):
"""ODIR-5K: one row per patient, per-eye diagnostic keywords separated by Chinese or ASCII commas."""
df = _read_table(src_dir / cfg["file"])
rows = []
for side in ("Left", "Right"):
fcol, kcol = f"{side}-Fundus", f"{side}-Diagnostic Keywords"
_need(df, ["ID", fcol, kcol], cfg["file"])
for _, r in df.iterrows():
kws = [k.strip() for k in re.split(r"[，,]", _norm(r[kcol])) if k.strip()]
rows.append([str(Path(cfg["image_dir"]) / _norm(r[fcol])), "|".join(kws), _norm(r["ID"]), side.lower(), ""])
return pd.DataFrame(rows, columns=COLS)
def adapter_labels_csv(src_dir, cfg):
"""Manually converted sources: <source>/labels.csv with image,label[,patient,eye,grade]; label tokens '|'-separated."""
df = _read_table(src_dir / cfg.get("file", "labels.csv"))
_need(df, ["image", "label"], "labels.csv")
get = lambda r, c: _norm(r[c]) if c in df.columns else ""
rows = [[get(r, "image"), get(r, "label"), get(r, "patient"), get(r, "eye").lower(), get(r, "grade")]
for _, r in df.iterrows()]
return pd.DataFrame(rows, columns=COLS)
ADAPTERS = {
"csv": adapter_csv,
"folders": adapter_folders,
"filename": adapter_filename,
"odir": adapter_odir,
"labels_csv": adapter_labels_csv,
}
def run_adapter(src, raw_root):
cfg = src["adapter"]
return ADAPTERS[cfg["type"]](Path(raw_root) / src["id"], cfg)
|