File size: 6,186 Bytes
b62601e | 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 | """
PyTorch Dataset for thermal image sequences.
Organises thermal images into fixed-length sequences for
training the CNN + LSTM pipeline. Supports train / val / test splits
and optional augmentation.
"""
import os
import random
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader
from pathlib import Path
from typing import List, Tuple, Optional
from src.preprocessing import ThermalImageProcessor, ThermalAugmentor
class ThermalSequenceDataset(Dataset):
"""
PyTorch Dataset that yields (sequence_tensor, label) pairs.
Directory layout expected:
data/sequences/
normal/
seq_001/
img_001.png
img_002.png
...
abnormal/
seq_010/
...
Each sequence folder contains chronologically ordered thermal images.
"""
LABEL_MAP = {"normal": 0, "abnormal": 1}
def __init__(
self,
sequences_dir: str,
processor: ThermalImageProcessor,
augmentor: Optional[ThermalAugmentor] = None,
sequence_length: int = 20,
split: str = "train",
train_ratio: float = 0.7,
val_ratio: float = 0.15,
seed: int = 42,
):
"""
Args:
sequences_dir: Root directory containing label sub-folders.
processor: ThermalImageProcessor instance.
augmentor: Optional ThermalAugmentor (applied only in train).
sequence_length: Fixed number of images per sequence.
split: One of 'train', 'val', 'test'.
train_ratio: Fraction of data for training.
val_ratio: Fraction of data for validation.
seed: Random seed for reproducible splits.
"""
super().__init__()
self.processor = processor
self.augmentor = augmentor if split == "train" else None
self.sequence_length = sequence_length
# Discover sequences
all_sequences = self._discover_sequences(sequences_dir)
random.seed(seed)
random.shuffle(all_sequences)
# Split
n = len(all_sequences)
n_train = int(n * train_ratio)
n_val = int(n * val_ratio)
if split == "train":
self.sequences = all_sequences[:n_train]
elif split == "val":
self.sequences = all_sequences[n_train : n_train + n_val]
else:
self.sequences = all_sequences[n_train + n_val :]
def _discover_sequences(self, root: str) -> List[Tuple[str, int]]:
"""
Walk the directory tree and return a list of
(sequence_folder_path, label) tuples.
"""
sequences = []
root = Path(root)
for label_name, label_id in self.LABEL_MAP.items():
label_dir = root / label_name
if not label_dir.exists():
continue
for seq_dir in sorted(label_dir.iterdir()):
if seq_dir.is_dir():
img_files = sorted([
str(f) for f in seq_dir.iterdir()
if f.suffix.lower() in (".png", ".jpg", ".jpeg", ".bmp", ".tiff")
])
if len(img_files) >= 1:
sequences.append((img_files, label_id))
return sequences
def _pad_or_truncate(self, images: List[np.ndarray]) -> List[np.ndarray]:
"""Ensure the sequence has exactly *sequence_length* frames."""
if len(images) >= self.sequence_length:
return images[: self.sequence_length]
# Pad by repeating the last frame
while len(images) < self.sequence_length:
images.append(images[-1].copy())
return images
def __len__(self) -> int:
return len(self.sequences)
def __getitem__(self, idx: int) -> Tuple[torch.Tensor, int]:
img_paths, label = self.sequences[idx]
# Load and preprocess
images = [self.processor.process(p) for p in img_paths]
images = self._pad_or_truncate(images)
# Augment (train only)
if self.augmentor is not None:
images = self.augmentor.augment_sequence(images)
# Stack → (T, 1, H, W)
tensors = [
torch.from_numpy(img).unsqueeze(0) for img in images
]
sequence_tensor = torch.stack(tensors, dim=0) # (T, 1, H, W)
return sequence_tensor, label
def create_dataloaders(
config,
processor: ThermalImageProcessor,
augmentor: Optional[ThermalAugmentor] = None,
) -> Tuple[DataLoader, DataLoader, DataLoader]:
"""
Build train / val / test DataLoaders from config.
Returns:
Tuple of (train_loader, val_loader, test_loader).
"""
seq_dir = config.data.sequences_dir
seq_len = config.data.sequence_length
batch = config.training.batch_size
workers = config.data.get("num_workers", 0)
datasets = {}
for split in ("train", "val", "test"):
datasets[split] = ThermalSequenceDataset(
sequences_dir=seq_dir,
processor=processor,
augmentor=augmentor if split == "train" else None,
sequence_length=seq_len,
split=split,
train_ratio=config.data.train_split,
val_ratio=config.data.val_split,
seed=config.get("seed", 42),
)
train_loader = DataLoader(
datasets["train"],
batch_size=batch,
shuffle=True,
num_workers=workers,
pin_memory=True,
drop_last=True,
)
val_loader = DataLoader(
datasets["val"],
batch_size=batch,
shuffle=False,
num_workers=workers,
pin_memory=True,
)
test_loader = DataLoader(
datasets["test"],
batch_size=batch,
shuffle=False,
num_workers=workers,
pin_memory=True,
)
return train_loader, val_loader, test_loader
|