-
Notifications
You must be signed in to change notification settings - Fork 4
/
dataloader_tf.py
240 lines (220 loc) · 10.8 KB
/
dataloader_tf.py
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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
CODE RELEASE TO SUPPORT RESEARCH.
COMMERCIAL USE IS NOT PERMITTED.
#==============================================================================
An implementation based on:
***
C.I. Nwoye, T. Yu, C. Gonzalez, B. Seeliger, P. Mascagni, D. Mutter, J. Marescaux, N. Padoy.
Rendezvous: Attention Mechanisms for the Recognition of Surgical Action Triplets in Endoscopic Videos.
Medical Image Analysis, 78 (2022) 102433.
***
Created on Thu Oct 21 15:38:36 2021
#==============================================================================
Copyright 2021 The Research Group CAMMA Authors All Rights Reserved.
(c) Research Group CAMMA, University of Strasbourg, France
@ Laboratory: CAMMA - ICube
@ Author: Chinedu Innocent Nwoye
@ Website: http://camma.u-strasbg.fr
#==============================================================================
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
#==============================================================================
"""
import os
import json
import random
import numpy as np
from PIL import Image
import tensorflow as tf
class CholecT50():
def __init__(self,
dataset_dir,
dataset_variant="cholect45-crossval",
test_fold=1,
augmentation_list=['original', 'vflip', 'hflip', 'contrast', 'rot90'],
normalize=True,
num_parallel_calls=8):
""" Args
dataset_dir : common path to the dataset (excluding videos, output)
list_video : list video IDs, e.g: ['VID01', 'VID02']
aug : data augumentation style
split : data split ['train', 'val', 'test']
Call
batch_size: int,
shuffle: True or False
Return
tuple ((image), (tool_label, verb_label, target_label, triplet_label))
"""
self.normalize = normalize
self.dataset_dir = dataset_dir
self.list_dataset_variant = {
"cholect45-crossval": "for CholecT45 dataset variant with the official cross-validation splits.",
"cholect50-crossval": "for CholecT50 dataset variant with the official cross-validation splits",
"cholect50-challenge": "for CholecT50 dataset variant as used in CholecTriplet challenge",
"cholect50": "for the CholecT50 dataset with original splits used in rendezvous paper",
"cholect45": "a pointer to cholect45-crossval",
}
assert dataset_variant in self.list_dataset_variant.keys(), print(dataset_variant, "is not a valid dataset variant")
video_split = self.split_selector(case=dataset_variant)
train_videos = sum([v for k,v in video_split.items() if k!=test_fold], []) if 'crossval' in dataset_variant else video_split['train']
test_videos = sum([v for k,v in video_split.items() if k==test_fold], []) if 'crossval' in dataset_variant else video_split['test']
if 'crossval' in dataset_variant:
val_videos = train_videos[-5:]
train_videos = train_videos[:-5]
else:
val_videos = video_split['val']
self.train_records = ['VID{}'.format(str(v).zfill(2)) for v in train_videos]
self.val_records = ['VID{}'.format(str(v).zfill(2)) for v in val_videos]
self.test_records = ['VID{}'.format(str(v).zfill(2)) for v in test_videos]
self.augmentation_list = augmentation_list
self.num_parallel_calls = self.get_num_parallel(num_parallel_calls)
self.build_train_dataset()
self.build_val_dataset()
self.build_test_dataset()
def split_selector(self, case='cholect50'):
switcher = {
'cholect50': {
'train': [1, 15, 26, 40, 52, 65, 79, 2, 18, 27, 43, 56, 66, 92, 4, 22, 31, 47, 57, 68, 96, 5, 23, 35, 48, 60, 70, 103, 13, 25, 36, 49, 62, 75, 110],
'val' : [8, 12, 29, 50, 78],
'test' : [6, 51, 10, 73, 14, 74, 32, 80, 42, 111]
},
'cholect50-challenge': {
'train': [1, 15, 26, 40, 52, 79, 2, 27, 43, 56, 66, 4, 22, 31, 47, 57, 68, 23, 35, 48, 60, 70, 13, 25, 49, 62, 75, 8, 12, 29, 50, 78, 6, 51, 10, 73, 14, 32, 80, 42],
'val': [5, 18, 36, 65, 74],
'test': [92, 96, 103, 110, 111]
},
'cholect45-crossval': {
1: [79, 2, 51, 6, 25, 14, 66, 23, 50,],
2: [80, 32, 5, 15, 40, 47, 26, 48, 70,],
3: [31, 57, 36, 18, 52, 68, 10, 8, 73,],
4: [42, 29, 60, 27, 65, 75, 22, 49, 12,],
5: [78, 43, 62, 35, 74, 1, 56, 4, 13,],
},
'cholect50-crossval': {
1: [79, 2, 51, 6, 25, 14, 66, 23, 50, 111],
2: [80, 32, 5, 15, 40, 47, 26, 48, 70, 96],
3: [31, 57, 36, 18, 52, 68, 10, 8, 73, 103],
4: [42, 29, 60, 27, 65, 75, 22, 49, 12, 110],
5: [78, 43, 62, 35, 74, 1, 56, 4, 13, 92],
},
}
return switcher.get(case)
def augmentation(self, img, label):
self.switcher_img = {
'original' : img,
'scaling' : self.scale(img),
'vflip' : tf.image.random_flip_up_down(img),
'hflip' : tf.image.random_flip_left_right(img),
'transpose' : self.transpose(img),
'rot90' : tf.image.rot90(img, k = 1),
'brightness' : tf.image.random_brightness(img, 0.5),
'contrast' : tf.image.random_contrast(img, 0.3, 0.5),
}
for case in self.augmentation_list:
img = self.switcher_img.get(case, img)
img = self.resize(img)
return img, label
def scale(self, img):
try:
rate = tf.random.uniform(shape=[2], minval=0.5, maxval=2.0, dtype=tf.float32, name='resize_rate')
r_shape = tf.cast(tf.cast(tf.shape(img)[1:3],tf.float32) * rate, tf.int32)
return tf.image.resize_with_crop_or_pad(img, r_shape[0], r_shape[1])
except:
rate = tf.random_uniform(shape=[2], minval=0.5, maxval=2.0, dtype=tf.float32, name='resize_rate')
r_shape = tf.cast(tf.cast(tf.shape(img)[1:3],tf.float32) * rate, tf.int32)
return tf.image.resize_image_with_crop_or_pad(img, r_shape[0], r_shape[1])
def transpose(self, img):
try:
return tf.image.transpose(img)
except:
return tf.image.transpose_image(img)
def resize(self, img):
try:
return tf.image.resize(images=img, size=[256, 448])
except:
return tf.image.resize_images(images=img, size=[256, 448])
def get_num_parallel(self, n):
try:
return tf.data.experimental.AUTOTUNE
except:
return n
def list_augmentations(self):
print(self.switcher_img.keys())
def list_dataset_variants(self):
print(self.list_dataset_variant)
def get_binary_labels(self, labels):
tool_label = np.zeros([6])
verb_label = np.zeros([10])
target_label = np.zeros([15])
triplet_label = np.zeros([100])
phase_label = np.zeros([100])
for label in labels:
triplet = label[0:1]
if triplet[0] != -1.0:
triplet_label[triplet[0]] = 1
tool = label[1:7]
if tool[0] != -1.0:
tool_label[tool[0]] = 1
verb = label[7:8]
if verb[0] != -1.0:
verb_label[verb[0]] = 1
target = label[8:14]
if target[0] != -1.0:
target_label[target[0]] = 1
phase = label[14:15]
if phase[0] != -1.0:
phase_label[phase[0]] = 1
return (triplet_label, tool_label, verb_label, target_label, phase_label)
def generator(self, records):
for record in records:
video_path = os.path.join(self.dataset_dir, "videos/{}".format(record.decode("utf-8")))
label_file = os.path.join(self.dataset_dir, "labels/{}.json".format(record.decode("utf-8")))
label_data = json.load(open(label_file, "rb"))
label_data = label_data["annotations"]
uframes = sorted(list(label_data.keys()))
for frame in uframes:
img_path = os.path.join(self.img_dir, "{:0>6}.png".format(self.frame))
image = Image.open(img_path)
image = image.resize(size=(448,256))
labels = label_data["annotations"][frame]
ivt, i, v, t, p = self.get_binary_labels(labels)
yield image, (ivt, i, v, t, p)
def build_train_dataset(self):
self.train_dataset = tf.data.Dataset.from_generator(
self.generator,
args = [self.train_records],
output_types = (tf.float32, (tf.int32, tf.int32, tf.int32, tf.int32, tf.int32)),
output_shapes = ([256, 448,3], ([6], [10], [15], [100], [7]))
)
self.train_dataset = self.train_dataset.map(self.augmentation, num_parallel_calls=self.num_parallel_calls)
def build_val_dataset(self):
self.val_dataset = tf.data.Dataset.from_generator(
self.generator,
args = [self.val_records],
output_types = (tf.float32, (tf.int32, tf.int32, tf.int32, tf.int32, tf.int32)),
output_shapes = ([256, 448,3], ([6], [10], [15], [100], [7]))
)
self.val_dataset = self.val_dataset.map(self.augmentation, num_parallel_calls=self.num_parallel_calls)
def build_test_dataset(self):
self.test_dataset = []
for video in self.test_records:
test_dataset = tf.data.Dataset.from_generator(
self.generator,
args = [[video]],
output_types = (tf.float32, (tf.int32, tf.int32, tf.int32, tf.int32, tf.int32)),
output_shapes = ([256, 448,3], ([6], [10], [15], [100], [7]))
)
self.test_dataset.append(test_dataset)
def build(self):
return (self.train_dataset, self.val_dataset, self.test_dataset)
if __name__ == "__main__":
print("Refers to https://github.com/CAMMA-public/cholect45 for the usage guide.")