def __getitem__(self, index):
path = self.paths[index]
# faster than niffty
#images, label = pkload(path + 'data_f32_divm.pkl')
#images, label = pkload(path + 'data_f32.pkl')
#images, label = torch.tensor(images), torch.tensor(label)
images, label = pkload(path + 'data_f32.pkl')
images, label = torch.from_numpy(images), torch.from_numpy(label)
if not self.crop:
# transformation needs nhwtc
images, label = images.unsqueeze(0), label.unsqueeze(0)
images, label = self.transforms([images, label])
images, label = images.squeeze(0), label.squeeze(0)
images = images.permute(3, 0, 1, 2).contiguous()
return (images, self.all_coords), label
if self.for_train:
fg, bg = pkload(path + self.suffix + 'coords.pkl')
coords = torch.cat(
[sample(x, self.num_patches // 2) for x in (fg, bg)])
else:
coords = self.all_coords
samples = multicrop.crop3d_cpu(images, coords, self.sample_size,
self.sample_size, self.sample_size, 1,
False)
sub_samples = multicrop.crop3d_cpu(images, coords,
self.sub_sample_size,
self.sub_sample_size,
self.sub_sample_size, 3, False)
if self.return_target:
target = multicrop.crop3d_cpu(label, coords, self.target_size,
self.target_size, self.target_size,
1, False)
samples, sub_samples, target = self.transforms(
[samples, sub_samples, target])
else:
samples, sub_samples = self.transforms([samples, sub_samples])
target = coords
if self.for_train: label = _zero
samples = samples.permute(0, 4, 1, 2, 3).contiguous()
sub_samples = sub_samples.permute(0, 4, 1, 2, 3).contiguous()
return (samples, sub_samples, target), label
import numpy as np
import torch
import multicrop
x = torch.randint(0, 5, (10, 15, 10, 3), dtype=torch.int16)
_stride = 5
_shape = (10, 15, 10)
coords = torch.tensor(np.stack([
v.reshape(-1) for v in np.meshgrid(
*[_stride // 2 + np.arange(0, s, _stride) for s in _shape],
indexing='ij')
], -1),
dtype=torch.int16)
#x = x.cuda()
#coords = coords.cuda()
y = multicrop.crop3d_cpu(x, coords, 5, 5, 5, 1, False)
#for t in y:
# print('='*10)
# print(t)
y = y.view(2, 3, 2, 5, 5, 5, 3).permute(0, 3, 1, 4, 2, 5,
6).reshape(10, 15, 10, 3)
print((x == y).all().item())
