|
a |
|
b/utils/dataloader.py |
|
|
1 |
import pickle |
|
|
2 |
import numpy as np |
|
|
3 |
import os |
|
|
4 |
from utils.equalizer import * |
|
|
5 |
from config import * |
|
|
6 |
|
|
|
7 |
#dataset_path: path to *.np dataset that has been preprocessed |
|
|
8 |
#normdata_path: path to directory where norm and equalization data is stored ('normalization.np','equalization.np') |
|
|
9 |
#img_res: the shape of the orthotopres (cubes) being processed |
|
|
10 |
class DataLoader(): |
|
|
11 |
def __init__(self, dataset_path, normdata_path, img_res=None): |
|
|
12 |
self.normdata_path = normdata_path |
|
|
13 |
if img_res is not None: |
|
|
14 |
self.img_res = img_res |
|
|
15 |
else: |
|
|
16 |
self.img_res = config['cube_size'] |
|
|
17 |
|
|
|
18 |
self.m_xlims = config['mask_xlims'] |
|
|
19 |
self.m_ylims = config['mask_ylims'] |
|
|
20 |
self.m_zlims = config['mask_zlims'] |
|
|
21 |
print("loading preprocessed dataset...") |
|
|
22 |
self.data_train = np.load(dataset_path) |
|
|
23 |
# format for nerual net |
|
|
24 |
self.data_train = self.data_train.reshape((len(self.data_train), self.img_res[0], self.img_res[1], self.img_res[2], 1)) |
|
|
25 |
# shuffle |
|
|
26 |
np.random.shuffle(self.data_train) |
|
|
27 |
|
|
|
28 |
def load_data(self, batch_size=1, is_testing=False): |
|
|
29 |
if is_testing == False: |
|
|
30 |
idx = np.random.permutation(len(self.data_train)) |
|
|
31 |
batch_images = self.data_train[idx[:batch_size]] |
|
|
32 |
else: |
|
|
33 |
idx = np.random.permutation(len(self.data_train)) |
|
|
34 |
batch_images = self.data_train[idx[:batch_size]] |
|
|
35 |
imgs_A = [] |
|
|
36 |
imgs_B = [] |
|
|
37 |
for i, img in enumerate(batch_images): |
|
|
38 |
imgs_A.append(img) |
|
|
39 |
img_out = np.copy(img) |
|
|
40 |
img_out[self.m_zlims[0]:self.m_zlims[1], self.m_xlims[0]:self.m_xlims[1], self.m_ylims[0]:self.m_ylims[1]] = 0 |
|
|
41 |
imgs_B.append(img_out) |
|
|
42 |
|
|
|
43 |
return np.array(imgs_A), np.array(imgs_B) |
|
|
44 |
|
|
|
45 |
def load_batch(self, batch_size=1, is_testing=False): |
|
|
46 |
if is_testing == False: |
|
|
47 |
self.n_batches = int(len(self.data_train) / batch_size) |
|
|
48 |
else: |
|
|
49 |
self.n_batches = int(len(self.data_train) / batch_size) |
|
|
50 |
|
|
|
51 |
for i in range(self.n_batches - 1): |
|
|
52 |
if is_testing == False: |
|
|
53 |
batch = self.data_train[i * batch_size:(i + 1) * batch_size] |
|
|
54 |
else: |
|
|
55 |
batch = self.data_train[i * batch_size:(i + 1) * batch_size] |
|
|
56 |
imgs_A = [] |
|
|
57 |
imgs_B = [] |
|
|
58 |
for i, img in enumerate(batch): |
|
|
59 |
imgs_A.append(img) |
|
|
60 |
img_out = np.copy(img) |
|
|
61 |
img_out[self.m_zlims[0]:self.m_zlims[1], self.m_xlims[0]:self.m_xlims[1], self.m_ylims[0]:self.m_ylims[1]] = 0 |
|
|
62 |
imgs_B.append(img_out) |
|
|
63 |
imgs_A = np.array(imgs_A) |
|
|
64 |
imgs_B = np.array(imgs_B) |
|
|
65 |
yield imgs_A, imgs_B |
|
|
66 |
|