-
Notifications
You must be signed in to change notification settings - Fork 0
/
mrcnn_trainer.py
340 lines (313 loc) · 12.4 KB
/
mrcnn_trainer.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
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
# PARAMETERS
learning_rate = 0.001
batch_size = 10
epochs = 2
train_iterations = 100
use_gpu = True
tensorflow_debug = False
model_path = 'mask_rcnn_coco.h5'
import logging
encoding = "utf-8-sig" # or utf-8
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
handler = logging.FileHandler('logger.log', encoding=encoding)
handler.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
handler.setFormatter(formatter)
logger.addHandler(handler)
logger.info('-' * 20 + 'START OF RUN' + '-' * 20)
# -*- coding: utf-8 -*-
logger.info('Loading modules')
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.allow_growth = True # Stops tf allocating all memory
session = tf.Session(config=config)
# Sets keras backend to tensorflow
import os
os.environ['KERAS_BACKEND'] = 'tensorflow'
if not use_gpu:
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
if not tensorflow_debug:
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
import cv2
import numpy as np
import glob
import keras
import json
import os
import random
from keras.models import Sequential, load_model
from keras.layers import Dense, Dropout, Flatten, Embedding, Activation, BatchNormalization
from keras.layers import Convolution1D, Convolution2D, MaxPooling2D, AveragePooling2D, LSTM
from keras.layers import GlobalAveragePooling2D, UpSampling2D, TimeDistributed, Reshape
from keras.callbacks import ModelCheckpoint, TensorBoard
from tensorflow.python.client import device_lib
from keras.utils import np_utils
from keras.optimizers import SGD, Adam, RMSprop
import pdb
import time
import matplotlib
import re
import multiprocessing
from multiprocessing import freeze_support, RLock
import itertools
import mrcnn
from mrcnn import utils
import mrcnn.model as modellib
from mrcnn import visualize
from mrcnn.config import Config as mrcnn_config
import tqdm
import pickle
from sklearn.model_selection import train_test_split
import imgaug
# Needed to change plot position while calculating. NEEDS TO ADDED BEFORE pyplot import
matplotlib.use('TkAgg')
import matplotlib.pyplot as plt
# https://www.pyimagesearch.com/2019/06/10/keras-mask-r-cnn/
# https://www.analyticsvidhya.com/blog/2019/07/computer-vision-implementing-mask-r-cnn-image-segmentation/
CLASS_NAMES = ['bg',
'building',
'road',
'water']
COLORS = [[0, 0, 0],
[255, 255, 0],
[255, 0, 255],
[255, 0, 0]]
"""
0, 0, 0 = Empty = Nothing = 0
255, 255, 0 = Yellow = building = 1
255, 0, 255 = Purple = road = 2
255, 0, 0 = Red = water = 3
"""
IDS = [0,1,2,3]
class Config(mrcnn_config):
NAME = "aerial_drone_map"
GPU_COUNT = 1
IMAGES_PER_GPU = 1 # batch size
NUM_CLASSES = len(CLASS_NAMES) # Needs to include bg
IMAGE_MIN_DIM = 1024
IMAGE_MAX_DIM = 1024
STEPS_PER_EPOCH = 500
IMAGES_PER_GPU = 1
# Use smaller anchors because our image and objects are small
# NOTE Different anchor sizes are suitable for houses/water/roads.
RPN_ANCHOR_SCALES = (64, 128, 256, 512, 1024) # anchor side in pixels
# Aim to allow ROI sampling to pick 33% positive ROIs.
TRAIN_ROIS_PER_IMAGE = 100
VALIDATION_STEPS = 25
class DroneDataset(utils.Dataset):
def load_data(self, train_data):
self.train_data = train_data
self.image_id_to_path = {}
for idx, img in enumerate(train_data.keys()):
self.image_id_to_path[idx] = img
self.add_class("object", IDS[1], "building")
self.add_class("object", IDS[2], "road")
self.add_class("object", IDS[3], "water")
for idx, file_path in enumerate(train_data.keys()):
self.add_image("object",
image_id=file_path, # For some reason, this won't work
path=file_path)
#self.train_data[idx+1] = self.train_data.pop(file_path)
def load_mask(self, image_id):
"""Generate instance masks for an image.
Returns:
masks: A bool array of shape [height, width, instance count] with
one mask per instance.
class_ids: a 1D array of class IDs of the instance masks.
"""
path = self.image_id_to_path[image_id]
mask = self.train_data[path][0]
class_ids = self.train_data[path][1]
return mask, class_ids
def image_reference(self, image_id): # Needed
"""Return the path of the image."""
return self.image_id_to_path[image_id]
def log_func(txt):
logger.info(txt)
def main():
if not os.path.isfile('preprocessed_data.pkl'):
all_groups = group_images()
random.shuffle(all_groups) # Shuffle data sets
pool = multiprocessing.Pool(multiprocessing.cpu_count(),
initializer=tqdm.tqdm.set_lock,
initargs=(RLock(),))
# NOTE: bad way of doing it. Spikes at 32 Gb or RAM. Perhaps a generator would be better?
logger.info('Preprocessing data')
all_groups = list_chunk(all_groups, multiprocessing.cpu_count())
id_and_groups = [(id+1, group) for id, group in enumerate(all_groups)]
train_data = pool.map(preprocess_data, id_and_groups)
pool.close()
train_data = merge_dicts(train_data)
with open('preprocessed_data.pkl', 'wb') as f:
pickle.dump(train_data, f)
#train_data = preprocess_data([1, all_groups])
else:
with open('preprocessed_data.pkl', 'rb') as f:
train_data = pickle.load(f)
train_data, val_data = dict_split(train_data, 0.7)
logger.info(f'Training: {len(list(train_data.keys()))} images, Validation: {len(list(val_data.keys()))} images')
dataset_train = prepare_dataset(train_data)
dataset_val = prepare_dataset(val_data)
model, predict_model, config = create_model()
fit_model(model, predict_model, config, dataset_train, dataset_train)
def prepare_dataset(data_dict):
dataset = DroneDataset()
dataset.load_data(data_dict)
dataset.prepare()
return dataset
def merge_dicts(dict_list):
first_dict = dict_list[0]
for d in dict_list[1:]:
first_dict.update(d)
return first_dict
def create_model():
config = Config()
config.display()
model = modellib.MaskRCNN(mode="training",
model_dir=os.getcwd(),
config=config)
model.keras_model.metrics_tensors = []
model.load_weights(model_path,
by_name=True,
exclude=[ "mrcnn_class_logits", "mrcnn_bbox_fc", "mrcnn_bbox", "mrcnn_mask"])
predict_model = modellib.MaskRCNN(mode="inference", config=config, # Will be used to predict
model_dir=os.getcwd())
predict_model.keras_model.metrics_tensors = []
return model, predict_model, config
def list_chunk(l, n_chunks):
avg = len(l) / float(n_chunks)
out = []
last = 0.0
while last < len(l):
out.append(l[int(last):int(last + avg)])
last += avg
return out
def dict_split(d, percent_split):
keys = list(d.keys())
new_dict = {}
for idx, key in enumerate(keys):
new_dict[key] = d[key]
del d[key]
if idx >= len(keys)*percent_split:
break
return new_dict, d
def preprocess_data(id_and_groups):
id = id_and_groups[0]
all_groups = id_and_groups[1]
progress = tqdm.tqdm(total=len(all_groups),
desc=f'Preprocessing data id {id}',
position=id)
data_dict = {}
for idx, group in enumerate(all_groups):
training_data = []
img, full_mask, building_mask, road_mask, water_mask, percent = group
img_array = img_to_array(img)
if img_array.shape != (1024, 1024, 3):
# Note: Perhaps we could make it so that the CNN takes any size, but the output size would be all over the place
#logger.warning(f'Image {img} has shape {img_array.shape}, not shape (1024, 1024, 3). Skipping')
continue
categorical_mask, categories = mask_to_categorical_array(img_array, full_mask)
data_dict[img] = [categorical_mask, categories]
progress.update(1)
return data_dict
def mask_to_categorical_array(img_array, mask):
if mask is None:
mask = np.zeros(img_array.shape)
return mask
mask = img_to_array(mask)
if mask.shape != img_array.shape:
logger.error("Wrong shape!!")
quit()
categories = [0]*(len(CLASS_NAMES)-1)
"""
NOTE: What you should do is create an empty matrix with (Width, Height, Categories)
However, the mask image is already (Widht, Height, Channels) <-- channels = colors,
meaning we don't need to create another one, so we just overwrite current mask.
This is just a funny, but useful coincidence in this case.
"""
for x in range(0, img_array.shape[0]):
for y in range(0, img_array.shape[1]):
channels_xy = mask[y,x]
if all(channels_xy == COLORS[0]): # BG
pixel = [False, False, False]
elif all(channels_xy == COLORS[1]): # building
pixel = [True, False, False]
if not IDS[1] in categories:
categories[0] = 1
elif all(channels_xy == COLORS[2]): # road
pixel = [False, True, False]
if not IDS[2] in categories:
categories[1] = 1
elif all(channels_xy == COLORS[3]): # water
pixel = [False, False, True]
if not IDS[3] in categories:
categories[2] = 1
else:
print("Whoa, how the hell did you get here???")
mask[y,x] = pixel #img_array[y,x]
categories = np.array(categories, dtype=np.int32) # Convert to array, or else it crashes
return mask, categories
def img_to_array(img_file):
img = cv2.imread(img_file) # Preprocess image as well and simplify?
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
return img
def fit_model(model, predict_model, config, dataset_train, dataset_val):
augmentation = imgaug.augmenters.Fliplr(0.5)
try:
# Training - Stage 1
model.train(dataset_train, dataset_val,
epochs=40,
learning_rate=config.LEARNING_RATE,
layers='heads',
augmentation=augmentation)
# Training - Stage 2
# Finetune layers from ResNet stage 4 and up
model.train(dataset_train, dataset_val,
learning_rate=config.LEARNING_RATE,
epochs=120,
layers='4+',
augmentation=augmentation)
# Training - Stage 3
# Fine tune all layers
print("Fine tune all layers")
model.train(dataset_train, dataset_val,
learning_rate=config.LEARNING_RATE / 10,
epochs=160,
layers='all',
augmentation=augmentation)
except KeyboardInterrupt:
logger.info("KeyboardInterrupt. Quitting")
quit()
def predict_to_mask(mask):
mask = mask*255
return mask
def group_images():
logger.info('Grouping images')
images = glob.glob('Training_dataset/Images/*')
all_masks = glob.glob('Training_dataset/Masks/all/*')
building_masks = glob.glob('Training_dataset/Masks/building/*')
road_masks = glob.glob('Training_dataset/Masks/road/*')
water_masks = glob.glob('Training_dataset/Masks/water/*')
percentages = glob.glob('Training_dataset/Percentages/*')
all_groups = []
for idx, img in enumerate(images):
img_basename = os.path.basename(img).split('.')[0]
full_mask = group_item_and_image(img_basename, all_masks)
building_mask = group_item_and_image(img_basename, building_masks)
road_mask = group_item_and_image(img_basename, road_masks)
water_mask = group_item_and_image(img_basename, water_masks)
percent = group_item_and_image(img_basename, percentages)
group = [img, full_mask, building_mask, road_mask, water_mask, percent]
all_groups.append(group)
print(f'{idx} done out ot {len(images)}', end='\r')
return all_groups
def group_item_and_image(img_basename, mask_list):
for mask in mask_list:
mask_basename = os.path.basename(mask).split('.')[0]
if mask_basename == img_basename:
return mask
return None
if __name__ == '__main__':
freeze_support() # Windows support
main()