File size: 2,879 Bytes
b5241eb |
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 |
import torch
from torch.utils.data import Dataset
import numpy as np
from pathlib import Path
from typing import List, Tuple, Optional
# Import your existing loaders
from load_volumes import process_fluent_export_sparse, VolumeSpec
from load_planes import process_plane_export, PlaneSpec
class WAKESETDataset(Dataset):
def __init__(self, root_dir: str, subset: str = 'train', augment: bool = False):
"""
Args:
root_dir: Path to WAKESET folder.
subset: 'train', 'val', or 'test'.
augment: If True, applies the rotation/flipping described in the paper.
"""
self.root = Path(root_dir) / "Volumes"
self.files = sorted(list(self.root.glob("Forward_*_CUBE_128.csv")))
self.augment = augment
# Simple split logic (matches paper Section 6.1)
# In reality, you might load a specific split file here
n = len(self.files)
if subset == 'train': self.files = self.files[:int(0.8*n)]
elif subset == 'val': self.files = self.files[int(0.8*n):int(0.9*n)]
else: self.files = self.files[int(0.9*n):]
def __len__(self):
# If augmenting, we implicitly have 4x data (handled via index modulo)
return len(self.files) * 4 if self.augment else len(self.files)
def __getitem__(self, idx):
# Handle Augmentation Indexing
if self.augment:
file_idx = idx // 4
aug_mode = idx % 4 # 0: None, 1: Flip, 2: Rot+, 3: Rot-
else:
file_idx = idx
aug_mode = 0
# Load Raw Data (Cached .npz preferred)
path = self.files[file_idx]
npz_path = path.with_suffix('.npz')
if npz_path.exists():
data = np.load(npz_path)
vol = data['velocity_magnitude'] # Shape (128, 128, 128)
else:
# Fallback to robust loader
raw = process_fluent_export_sparse(path, fill_value=0.0)
vol = raw['velocity_magnitude']
# Convert to Tensor
tensor = torch.from_numpy(vol).float().unsqueeze(0) # (C, D, H, W)
# Apply Physics-Consistent Augmentation (Paper Section 5.3)
if aug_mode == 1:
# Flip across vertical mid-plane (assumes symmetry at 0-deg)
tensor = torch.flip(tensor, dims=[2]) # Flip Y-axis
# Note: True rotation requires rotating vector components (u,v,w)
# not just the scalar magnitude grid.
# Extract Kinematics from filename for Conditioning
# "Forward_0100_ms..."
spec = VolumeSpec.from_filename(path)
speed = float(spec.velocity) / 1000.0
angle = float(spec.angle)
kinematics = torch.tensor([speed, angle], dtype=torch.float32)
return tensor, kinematics |