-
Notifications
You must be signed in to change notification settings - Fork 31
/
trainorig.py
352 lines (302 loc) · 14.2 KB
/
trainorig.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
341
342
343
344
345
346
347
348
349
350
351
352
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import optparse
import os, sys
import numpy as np
import pandas as pd
from PIL import Image
import torch
from torch.optim.lr_scheduler import _LRScheduler
from torch.optim.lr_scheduler import ReduceLROnPlateau
import torch.nn as nn
import torch.nn.functional as F
from sklearn.model_selection import KFold
import os
import cv2
import glob
import numpy as np
import pandas as pd
import torch.optim as optim
from albumentations import Compose, ShiftScaleRotate, Resize
from albumentations.pytorch import ToTensor
from torch.utils.data import Dataset
from sklearn.metrics import log_loss
from torch.utils.data import DataLoader
import cv2
import gc
import random
import logging
import datetime
import torchvision
from torchvision import transforms as T
from torchvision.models.resnet import ResNet, Bottleneck
from torch.hub import load_state_dict_from_url
from torchvision.models.resnet import ResNet, Bottleneck
from albumentations import (Cutout, Compose, Normalize, RandomRotate90, HorizontalFlip,
VerticalFlip, ShiftScaleRotate, Transpose, OneOf, IAAAdditiveGaussianNoise,
GaussNoise, RandomGamma, RandomContrast, RandomBrightness, HueSaturationValue,
RandomBrightnessContrast, Lambda, NoOp, CenterCrop, Resize
)
from tqdm import tqdm
from apex import amp
from apex.parallel import DistributedDataParallel as DDP
from apex.fp16_utils import *
from apex import amp, optimizers
from apex.multi_tensor_apply import multi_tensor_applier
import warnings
warnings.filterwarnings('ignore')
# Print info about environments
parser = optparse.OptionParser()
parser.add_option('-s', '--seed', action="store", dest="seed", help="model seed", default="1234")
parser.add_option('-o', '--fold', action="store", dest="fold", help="Fold for split", default="0")
parser.add_option('-p', '--nbags', action="store", dest="nbags", help="Number of bags for averaging", default="0")
parser.add_option('-e', '--epochs', action="store", dest="epochs", help="epochs", default="5")
parser.add_option('-j', '--start', action="store", dest="start", help="Start epochs", default="0")
parser.add_option('-b', '--batchsize', action="store", dest="batchsize", help="batch size", default="16")
parser.add_option('-r', '--rootpath', action="store", dest="rootpath", help="root directory", default="/share/dhanley2/rsna/")
parser.add_option('-i', '--imgpath', action="store", dest="imgpath", help="root directory", default="data/mount/512X512X6/")
parser.add_option('-w', '--workpath', action="store", dest="workpath", help="Working path", default="densenetv1/weights")
parser.add_option('-f', '--weightsname', action="store", dest="weightsname", help="Weights file name", default="pytorch_model.bin")
parser.add_option('-l', '--lr', action="store", dest="lr", help="learning rate", default="0.00005")
parser.add_option('-g', '--logmsg', action="store", dest="logmsg", help="root directory", default="Recursion-pytorch")
parser.add_option('-c', '--size', action="store", dest="size", help="model size", default="512")
parser.add_option('-a', '--infer', action="store", dest="infer", help="root directory", default="TRN")
parser.add_option('-z', '--wtsize', action="store", dest="wtsize", help="model size", default="999")
parser.add_option('-m', '--hflip', action="store", dest="hflip", help="Augmentation - Embedding horizontal flip", default="F")
parser.add_option('-d', '--transpose', action="store", dest="transpose", help="Augmentation - Embedding transpose", default="F")
parser.add_option('-x', '--stage2', action="store", dest="stage2", help="Stage2 embeddings only", default="F")
parser.add_option('-y', '--autocrop', action="store", dest="autocrop", help="Autocrop", default="T")
options, args = parser.parse_args()
package_dir = options.rootpath
sys.path.append(package_dir)
sys.path.insert(0, 'scripts')
from logs import get_logger
from utils import dumpobj, loadobj, GradualWarmupScheduler
# Print info about environments
logger = get_logger(options.logmsg, 'INFO') # noqa
logger.info('Cuda set up : time {}'.format(datetime.datetime.now().time()))
device=torch.device('cuda')
logger.info('Device : {}'.format(torch.cuda.get_device_name(0)))
logger.info('Cuda available : {}'.format(torch.cuda.is_available()))
n_gpu = torch.cuda.device_count()
logger.info('Cuda n_gpus : {}'.format(n_gpu ))
logger.info('Load params : time {}'.format(datetime.datetime.now().time()))
for (k,v) in options.__dict__.items():
logger.info('{}{}'.format(k.ljust(20), v))
SEED = int(options.seed)
SIZE = int(options.size)
WTSIZE=int(options.wtsize) if int(options.wtsize) != 999 else SIZE
EPOCHS = int(options.epochs)
START = int(options.start)
AUTOCROP=options.autocrop=='T'
n_epochs = EPOCHS
lr=float(options.lr)
batch_size = int(options.batchsize)
ROOT = options.rootpath
path_data = os.path.join(ROOT, 'data')
path_img = os.path.join(ROOT, options.imgpath)
WORK_DIR = os.path.join(ROOT, options.workpath)
WEIGHTS_NAME = options.weightsname
fold = int(options.fold)
INFER=options.infer
HFLIP = 'T' if options.hflip=='T' else ''
TRANSPOSE = 'P' if options.transpose=='T' else ''
STAGE2=options.stage2=='T'
#classes = 1109
device = 'cuda'
print('Data path : {}'.format(path_data))
print('Image path : {}'.format(path_img))
os.environ["TORCH_HOME"] = os.path.join( path_data, 'mount')
logger.info(os.system('$TORCH_HOME'))
class Identity(nn.Module):
def __init__(self):
super(Identity, self).__init__()
def forward(self, x):
return x
def autocrop(image, threshold=0):
"""Crops any edges below or equal to threshold
Crops blank image to 1x1.
Returns cropped image.
https://stackoverflow.com/questions/13538748/crop-black-edges-with-opencv
"""
if len(image.shape) == 3:
flatImage = np.max(image, 2)
else:
flatImage = image
rows = np.where(np.max(flatImage, 0) > threshold)[0]
cols = np.where(np.max(flatImage, 1) > threshold)[0]
image = image[cols[0]: cols[-1] + 1, rows[0]: rows[-1] + 1]
#logger.info(image.shape)
sqside = max(image.shape)
imageout = np.zeros((sqside, sqside, 3), dtype = 'uint8')
imageout[:image.shape[0], :image.shape[1],:] = image.copy()
return imageout
class IntracranialDataset(Dataset):
def __init__(self, df, path, labels, transform=None):
self.path = path
self.data = df
self.transform = transform
self.labels = labels
self.crop = AUTOCROP
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
img_name = os.path.join(self.path, self.data.loc[idx, 'Image'] + '.jpg')
#img = cv2.imread(img_name, cv2.IMREAD_GRAYSCALE)
img = cv2.imread(img_name)
if self.crop:
try:
try:
img = autocrop(img, threshold=0, kernsel_size = image.shape[0]//15)
except:
img = autocrop(img, threshold=0)
except:
1
img = cv2.resize(img,(SIZE,SIZE))
if self.transform:
augmented = self.transform(image=img)
img = augmented['image']
if self.labels:
labels = torch.tensor(
self.data.loc[idx, label_cols])
return {'image': img, 'labels': labels}
else:
return {'image': img}
np.random.seed(SEED)
torch.manual_seed(SEED)
torch.cuda.manual_seed(SEED)
if n_gpu > 0:
torch.cuda.manual_seed_all(SEED)
torch.backends.cudnn.deterministic = True
logger.info('Load Dataframes')
dir_train_img = os.path.join(path_data, 'proc')
dir_test_img = os.path.join(path_data, 'proc')
# Parameters
n_classes = 6
label_cols = ['epidural', 'intraparenchymal', 'intraventricular', 'subarachnoid', 'subdural', 'any']
train = pd.read_csv(os.path.join(path_data, 'train.csv.gz'))
test = pd.read_csv(os.path.join(path_data, 'test.csv.gz'))
png = glob.glob(os.path.join(dir_train_img, '*.jpg'))
png = [os.path.basename(png)[:-4] for png in png]
train_imgs = set(train.Image.tolist())
png = [p for p in png if p in train_imgs]
logger.info('Number of images to train on {}'.format(len(png)))
png = np.array(png)
train = train.set_index('Image').loc[png].reset_index()
# get fold
valdf = train[train['fold']==fold].reset_index(drop=True)
trndf = train[train['fold']!=fold].reset_index(drop=True)
# Data loaders
mean_img = [0.22363983, 0.18190407, 0.2523437 ]
std_img = [0.32451536, 0.2956294, 0.31335256]
transform_train = Compose([
HorizontalFlip(p=0.5),
ShiftScaleRotate(shift_limit=0.05, scale_limit=0.05,
rotate_limit=20, p=0.3, border_mode = cv2.BORDER_REPLICATE),
Transpose(p=0.5),
Normalize(mean=mean_img, std=std_img, max_pixel_value=255.0, p=1.0),
ToTensor()
])
HFLIPVAL = 1.0 if HFLIP == 'T' else 0.0
TRANSPOSEVAL = 1.0 if TRANSPOSE == 'P' else 0.0
transform_test= Compose([
HorizontalFlip(p=HFLIPVAL),
Transpose(p=TRANSPOSEVAL),
Normalize(mean=mean_img, std=std_img, max_pixel_value=255.0, p=1.0),
ToTensor()
])
trndataset = IntracranialDataset(trndf, path=dir_train_img, transform=transform_train, labels=True)
valdataset = IntracranialDataset(valdf, path=dir_train_img, transform=transform_test, labels=False)
tstdataset = IntracranialDataset(test, path=dir_test_img, transform=transform_test, labels=False)
num_workers = 16
trnloader = DataLoader(trndataset, batch_size=batch_size, shuffle=True, num_workers=num_workers)
valloader = DataLoader(valdataset, batch_size=batch_size*4, shuffle=False, num_workers=num_workers)
tstloader = DataLoader(tstdataset, batch_size=batch_size*4, shuffle=False, num_workers=num_workers)
model = torch.load('checkpoints/resnext101_32x8d_wsl_checkpoint.pth')
model.fc = torch.nn.Linear(2048, n_classes)
model.to(device)
criterion = torch.nn.BCEWithLogitsLoss()
def criterion(data, targets, criterion = torch.nn.BCEWithLogitsLoss()):
''' Define custom loss function for weighted BCE on 'target' column '''
loss_all = criterion(data, targets)
loss_any = criterion(data[:,-1:], targets[:,-1:])
return (loss_all*6 + loss_any*1)/7
plist = [{'params': model.parameters(), 'lr': lr}]
optimizer = optim.Adam(plist, lr=lr)
model, optimizer = amp.initialize(model, optimizer, opt_level="O1")
model = torch.nn.DataParallel(model, device_ids=list(range(n_gpu)))
for epoch in range(n_epochs):
logger.info('Epoch {}/{}'.format(epoch, n_epochs - 1))
logger.info('-' * 10)
if INFER == 'TRN':
for param in model.parameters():
param.requires_grad = True
model.train()
tr_loss = 0
for step, batch in enumerate(trnloader):
if step%1000==0:
logger.info('Train step {} of {}'.format(step, len(trnloader)))
inputs = batch["image"]
labels = batch["labels"]
inputs = inputs.to(device, dtype=torch.float)
labels = labels.to(device, dtype=torch.float)
outputs = model(inputs)
loss = criterion(outputs, labels)
with amp.scale_loss(loss, optimizer) as scaled_loss:
scaled_loss.backward()
tr_loss += loss.item()
optimizer.step()
optimizer.zero_grad()
del inputs, labels, outputs
epoch_loss = tr_loss / len(trnloader)
logger.info('Training Loss: {:.4f}'.format(epoch_loss))
for param in model.parameters():
param.requires_grad = False
output_model_file = 'weights/model_{}_epoch{}_fold{}.bin'.format(WTSIZE, epoch, fold)
torch.save(model.state_dict(), output_model_file)
else:
del model
#model = torch.hub.load('rwightman/gen-efficientnet-pytorch', 'efficientnet_b0', pretrained=True)
model = torch.load('checkpoints/resnext101_32x8d_wsl_checkpoint.pth')
model.fc = torch.nn.Linear(2048, n_classes)
device = torch.device("cuda:{}".format(n_gpu-1))
model.to(device)
model = torch.nn.DataParallel(model, device_ids=list(range(n_gpu)[::-1]), output_device=device)
for param in model.parameters():
param.requires_grad = False
input_model_file = os.path.join(WORK_DIR, 'weights/model_{}_epoch{}_fold{}.bin'.format(WTSIZE, epoch, fold))
model.load_state_dict(torch.load(input_model_file))
model.to(device)
model.eval()
logger.info(model.parameters())
if INFER=='EMB':
logger.info('Output embeddings epoch {}'.format(epoch))
logger.info('Train shape {} {}'.format(*trndf.shape))
logger.info('Valid shape {} {}'.format(*valdf.shape))
logger.info('Test shape {} {}'.format(*test.shape))
trndataset = IntracranialDataset(trndf, path=dir_train_img, transform=transform_test, labels=False)
valdataset = IntracranialDataset(valdf, path=dir_train_img, transform=transform_test, labels=False)
tstdataset = IntracranialDataset(test, path=dir_test_img, transform=transform_test, labels=False)
trnloader = DataLoader(trndataset, batch_size=batch_size*4, shuffle=False, num_workers=num_workers)
valloader = DataLoader(valdataset, batch_size=batch_size*4, shuffle=False, num_workers=num_workers)
tstloader = DataLoader(tstdataset, batch_size=batch_size*4, shuffle=False, num_workers=num_workers)
# Extract embedding layer
model.module.fc = Identity()
model.eval()
DATASETS = ['tst', 'val', 'trn']
LOADERS = [tstloader, valloader, trnloader]
for typ, loader in zip(DATASETS, LOADERS):
ls = []
for step, batch in enumerate(loader):
if step%1000==0:
logger.info('Embedding {} step {} of {}'.format(typ, step, len(loader)))
inputs = batch["image"]
inputs = inputs.to(device, dtype=torch.float)
out = model(inputs)
ls.append(out.detach().cpu().numpy())
outemb = np.concatenate(ls, 0).astype(np.float32)
logger.info('Write embeddings : shape {} {}'.format(*outemb))
np.savez_compressed(os.path.join(WORK_DIR, 'emb{}_{}_size{}_fold{}_ep{}'.format(HFLIP+TRANSPOSE, typ, SIZE, fold, epoch), outemb))
dumpobj(os.path.join(WORK_DIR, 'loader{}_{}_size{}_fold{}_ep{}'.format(HFLIP+TRANSPOSE, typ, SIZE, fold, epoch), loader))
gc.collect()