File size: 10,508 Bytes
dadf189 | 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 | from __future__ import annotations
"""Dataset for fingerprint images + pre-extracted minutiae.
Supports two directory layouts:
1. **Paired layout** (image_dir and minutiae_dir separate):
image_dir/<identity>/<sample>.{bmp,png,tif,jpg}
minutiae_dir/<identity>/<sample>.txt
2. **PolyU layout** (images in session dirs, minutiae in processed dir):
image_root/first_session/{finger_id}_{num}.jpg
minutiae_root/{split}/finger_{finger_id}/{session}_{finger_id}_{num}.txt
Each minutiae .txt file has one minutia per line: ``x y theta [type]``.
"""
import os
import random
from collections import defaultdict
from pathlib import Path
from typing import Any
import torch
from PIL import Image
from torch.utils.data import DataLoader, Dataset, Sampler
from torchvision import transforms
from ..configs.default import AugmentConfig, DataConfig
from .joint_augmentor import JointAugmentor
class FingerprintDataset(Dataset):
"""Loads fingerprint images paired with pre-extracted minutiae .txt files.
Returns dict:
image: (1, H, W) normalized [0, 1]
minutiae_raw: (max_N, 3) pixel [x, y, theta]
mask: (max_N,) bool
label: int
"""
IMAGE_EXTENSIONS = {".bmp", ".png", ".tif", ".tiff", ".jpg", ".jpeg"}
def __init__(
self,
image_dir: str,
minutiae_dir: str,
image_size: tuple[int, int] = (256, 256),
max_minutiae: int = 120,
min_minutiae: int = 10,
augment: bool = True,
augment_cfg: AugmentConfig | None = None,
):
super().__init__()
self.image_size = image_size
self.max_minutiae = max_minutiae
self.min_minutiae = min_minutiae
self.augmentor = (
JointAugmentor(augment_cfg or AugmentConfig()) if augment else None
)
self.to_tensor = transforms.ToTensor()
self.samples: list[tuple[str, str, int]] = [] # (img_path, txt_path, label)
self.label_map: dict[str, int] = {}
self._discover_samples(image_dir, minutiae_dir)
def _discover_samples(self, image_dir: str, minutiae_dir: str):
"""Scan directories and pair images with minutiae files."""
image_root = Path(image_dir)
minutiae_root = Path(minutiae_dir)
if not image_root.exists() or not minutiae_root.exists():
return
for idx, identity_dir in enumerate(sorted(minutiae_root.iterdir())):
if not identity_dir.is_dir():
continue
identity_name = identity_dir.name
self.label_map[identity_name] = idx
for txt_file in sorted(identity_dir.glob("*.txt")):
img_path = self._find_matching_image(
txt_file, image_root, identity_name
)
if img_path is not None:
self.samples.append((str(img_path), str(txt_file), idx))
def _find_matching_image(
self, txt_file: Path, image_root: Path, identity_name: str
) -> Path | None:
"""Find the image file corresponding to a minutiae .txt file.
Supports multiple dataset layouts:
- Paired: image_root/identity/sample.{ext}
- PolyU: image_root/{first,second}_session/finger_sample.{ext}
- FVC: image_root/FVC20XX/Dbs/DbN_{a,b}/finger_impr.{ext}
"""
stem = txt_file.stem
# Strategy 1: same directory structure (paired layout)
for ext in self.IMAGE_EXTENSIONS:
candidate = image_root / identity_name / f"{stem}{ext}"
if candidate.exists():
return candidate
# Strategy 2: PolyU layout — txt stem "s1_101_3" -> first_session/101_3.jpg
if stem.startswith(("s1_", "s2_")):
session_prefix = stem[:2]
original_name = stem[3:]
session_dir = (
"first_session" if session_prefix == "s1" else "second_session"
)
for ext in self.IMAGE_EXTENSIONS:
candidate = image_root / session_dir / f"{original_name}{ext}"
if candidate.exists():
return candidate
# Strategy 3: FVC layout — identity "fvc2000_db1_a_001", stem "1_3"
# -> image_root/FVC2000/Dbs/Db1_a/1_3.tif
if identity_name.startswith("fvc"):
parts = identity_name.split("_") # ["fvc2000", "db1", "a", "001"]
if len(parts) >= 3:
version = parts[0].upper().replace("FVC", "FVC") # "FVC2000"
db_set = f"{parts[1].capitalize()}_{parts[2]}" # "Db1_a"
for ext in self.IMAGE_EXTENSIONS:
candidate = image_root / version / "Dbs" / db_set / f"{stem}{ext}"
if candidate.exists():
return candidate
# Strategy 4: recursive search (fallback)
for ext in self.IMAGE_EXTENSIONS:
matches = list(image_root.rglob(f"{stem}{ext}"))
if matches:
return matches[0]
return None
@property
def num_classes(self) -> int:
return len(self.label_map)
@staticmethod
def _load_minutiae(path: str) -> torch.Tensor:
"""Read minutiae file -> (N, 3) tensor [x, y, theta]."""
rows = []
with open(path) as f:
for line in f:
parts = line.strip().split()
if len(parts) >= 3:
x, y, theta = float(parts[0]), float(parts[1]), float(parts[2])
rows.append([x, y, theta])
if len(rows) == 0:
return torch.zeros(1, 3)
return torch.tensor(rows, dtype=torch.float32)
def _scale_minutiae(
self,
minutiae: torch.Tensor,
orig_size: tuple[int, int],
) -> torch.Tensor:
"""Scale minutiae (x, y) from original image coords to resized coords."""
orig_h, orig_w = orig_size
target_h, target_w = self.image_size
m = minutiae.clone()
m[:, 0] *= target_w / orig_w
m[:, 1] *= target_h / orig_h
return m
def _pad_or_truncate(
self, minutiae: torch.Tensor
) -> tuple[torch.Tensor, torch.Tensor]:
"""Pad/truncate to max_minutiae; return (padded, mask)."""
N = minutiae.shape[0]
if N > self.max_minutiae:
perm = torch.randperm(N)[: self.max_minutiae]
minutiae = minutiae[perm]
N = self.max_minutiae
padded = torch.zeros(self.max_minutiae, 3)
padded[:N] = minutiae
mask = torch.zeros(self.max_minutiae, dtype=torch.bool)
mask[:N] = True
return padded, mask
def __len__(self) -> int:
return len(self.samples)
def __getitem__(self, idx: int, _depth: int = 0) -> dict[str, Any]:
img_path, txt_path, label = self.samples[idx]
# Load image
pil_img = Image.open(img_path).convert("L")
orig_w, orig_h = pil_img.size
pil_img = pil_img.resize(
(self.image_size[1], self.image_size[0]), Image.BILINEAR
)
image = self.to_tensor(pil_img)
# Load minutiae and scale to resized image coords
minutiae = self._load_minutiae(txt_path)
minutiae = self._scale_minutiae(minutiae, (orig_h, orig_w))
# Skip too-small samples
if minutiae.shape[0] < self.min_minutiae:
if _depth < len(self):
return self.__getitem__((idx + 1) % len(self), _depth + 1)
minutiae = torch.zeros(1, 3)
# Joint augmentation
if self.augmentor is not None:
image, minutiae = self.augmentor(image, minutiae)
# Pad/truncate
minutiae_raw, mask = self._pad_or_truncate(minutiae)
return {
"image": image,
"minutiae_raw": minutiae_raw,
"mask": mask,
"label": label,
}
class PKSampler(Sampler):
"""P identities x K samples per identity for metric learning."""
def __init__(self, dataset: FingerprintDataset, p: int = 8, k: int = 4):
self.dataset = dataset
self.p = p
self.k = k
self.label_to_indices: dict[int, list[int]] = defaultdict(list)
for idx, (_, _, label) in enumerate(dataset.samples):
self.label_to_indices[label].append(idx)
self.labels = [l for l, idxs in self.label_to_indices.items() if len(idxs) >= k]
if len(self.labels) < p:
self.labels = list(self.label_to_indices.keys())
def __iter__(self):
label_pool = self.labels.copy()
random.shuffle(label_pool)
batches_yielded = 0
target_batches = len(self)
ptr = 0
while batches_yielded < target_batches:
if ptr + self.p > len(label_pool):
label_pool = self.labels.copy()
random.shuffle(label_pool)
ptr = 0
batch_labels = label_pool[ptr : ptr + self.p]
ptr += self.p
batch = []
for label in batch_labels:
indices = self.label_to_indices[label]
if len(indices) >= self.k:
chosen = random.sample(indices, self.k)
else:
chosen = random.choices(indices, k=self.k)
batch.extend(chosen)
yield batch
batches_yielded += 1
def __len__(self):
return len(self.dataset) // (self.p * self.k)
def build_dataloader(
cfg: DataConfig,
split: str = "train",
batch_size: int = 32,
image_root: str | None = None,
) -> DataLoader:
"""Build a DataLoader for the model.
Args:
cfg: DataConfig with dirs, sizes, etc.
split: "train" or "val".
batch_size: batch size.
image_root: override for image directory root.
"""
augment = split == "train"
minutiae_dir = os.path.join(cfg.minutiae_dir, split)
img_dir = image_root or cfg.image_dir
ds = FingerprintDataset(
image_dir=img_dir,
minutiae_dir=minutiae_dir,
image_size=cfg.image_size,
max_minutiae=cfg.max_minutiae,
min_minutiae=cfg.min_minutiae,
augment=augment,
augment_cfg=cfg.augment if augment else None,
)
return DataLoader(
ds,
batch_size=batch_size,
shuffle=(split == "train"),
num_workers=cfg.num_workers,
pin_memory=cfg.pin_memory,
drop_last=(split == "train"),
)
|