label_species = int(self.file_info.iloc[idx]['species'])
label_classes = int(self.file_info.iloc[idx]['classes'])
sample = {
'image': image,
'species': label_species,
'classes': label_classes
}
if self.transform:
sample['image'] = self.transform(image)
return sample
train_transforms = transforms.Compose([
transforms.Resize((500, 500)),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
]) #?0.5??????????PIL??
val_transforms = transforms.Compose(
[transforms.Resize((500, 500)),
transforms.ToTensor()])
train_dataset = MyDataset(root_dir=ROOT_DIR + TRAIN_DIR,
annotations_file=TRAIN_ANNO,
transform=train_transforms)
test_dataset = MyDataset(root_dir=ROOT_DIR + VAL_DIR,
annotations_file=VAL_ANNO,
transform=val_transforms)
train_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)
def corrupted_cifar_uncertainty(method,
batch_size,
use_extra_corruptions=False,
dataset='cifar10',
normalize=False):
assert dataset in ['cifar100', 'cifar10']
if dataset == 'cifar10':
dataset = CorruptedCifar10
t = [
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465),
(0.2023, 0.1994, 0.2010))
]
else:
dataset = CorruptedCifar100
t = [
transforms.ToTensor(),
transforms.Normalize((0.5071, 0.4867, 0.4408),
(0.2675, 0.2565, 0.2761))
]
if use_extra_corruptions:
corruptions = chain(BENCHMARK_CORRUPTIONS, EXTRA_CORRUPTIONS)
else:
corruptions = BENCHMARK_CORRUPTIONS
scores = {name: {} for name in corruptions}
entropy = {name: {} for name in corruptions}
preds = {name: {} for name in corruptions}
eces = {name: {} for name in corruptions}
true_labels = None
for name in corruptions:
for severity in range(1, 6):
# CIFAR 10
loader = torch.utils.data.DataLoader(dataset=dataset(
'./datasets/',
download=True,
corruption=name,
severity=severity,
transform=transforms.Compose(t)),
batch_size=batch_size,
shuffle=False,
pin_memory=True,
num_workers=4)
ece, _, _, _ = ece_score(method, loader)
eces[name][severity] = ece
scores[name][severity] = eval_method(method, dataset=loader)
probs, labels, predictions = get_logits(method, loader)
if true_labels is None:
true_labels = labels
hs = []
for x, y in loader:
# true.extend(y.tolist())
p, _ = method.predict_proba(x, y, True)
plog = (p + 1e-12).log()
h = plog * p
if normalize:
h = h / np.log(h.shape[-1])
h = -torch.sum(h, -1)
hs.extend(h.tolist())
entropy[name][severity] = hs
preds[name][severity] = predictions
return entropy, preds, scores, true_labels
from sklearn.preprocessing import OneHotEncoder
#图片路径
train_list = sorted([
os.path.join('dataset1/train/', img)
for img in os.listdir('dataset1/train/')
])
trainGT_list = sorted([
os.path.join('dataset1/train_GT/SEG/', img)
for img in os.listdir('dataset1/train_GT/SEG/')
])
imgsize = 628
#???
data_transforms = transforms.Compose([
#transforms.RandomHorizontalFlip()
#transforms.Pad(92,padding_mode='symmetric'),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5])
])
datalabel_transforms = transforms.Compose([
transforms.ToTensor(),
])
#numpy形式
def loadpro_img(file_names):
images = []
for file_name in file_names:
img = cv2.imread(file_name, -1) #读取时不做改变，默认参数1加载彩色图片
img = img.astype(np.uint8)
#img = Image.fromarray(img)
images.append(img)
if True:
std = tch.exp(0.5 * logvar)
eps = tch.randn_like(std)
return eps.mul(std).add_(mu)
else:
return mu
def loss_function(recon_x, x, mu, logvar):
BCE = F.binary_cross_entropy(recon_x, x.view(-1, 784), size_average=False)
KLD = -0.5 * tch.sum(1 + logvar - mu.pow(2) - logvar.exp())
return BCE + KLD
tr = transforms.Compose([transforms.ToTensor()])
train = MNIST('./data', transform=tr, download=True)
test = MNIST('./data', transform=tr, train=False, download=True)
epoch = 10
batch_size = 30
train = DataLoader(train, batch_size, True)
test = DataLoader(test, batch_size, False)
vae = VAE((1, 28, 28), 12)
sgd = tch.optim.RMSprop(vae.parameters())
mse = nn.MSELoss(size_average=False)
for j in range(epoch):
con = 0
dis = 0
def gen_transform_validation(crop_size=448):
return transforms.Compose([
transforms.CenterCrop(crop_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5], std=[0.5])
])
def build_transforms(cfg):
return transforms.Compose(transforms.ToTensor())
import argparse
import pickle
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.models import resnet18
from torchvision.transforms import transforms
from tensorboardX import SummaryWriter
writer = SummaryWriter()
transform = transforms.Compose([
transforms.Resize(255),
transforms.RandomCrop(224),
transforms.ToTensor()
])
parser = argparse.ArgumentParser()
parser.add_argument('--dataset-dir',
type=str,
default='PKUMMDv1',
help='dataset directory')
parser.add_argument('--gpu',
default=False,
action='store_true',
help='whether to use gpus for training')
parser.add_argument('--batch-size', type=int, default=128, help='batch size')
parser.add_argument('--learning-rate',
type=float,
def main(args, cfgs):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
logging.info(f'Using device {device}')
############################################
class_names = cfgs["class_names"]
n_classes = cfgs["n_classes"]
multi_gpu = cfgs["gpus"]
Net = cfgs["net_name"]
image_root = cfgs["image_root"]
epochs = cfgs["epochs"]
batch_size_train = cfgs["batch_size_train"]
checkpoint_dir = cfgs["checkpoint_dir"]
checkpoint_space = cfgs["checkpoint_space"]
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
############################################
validate_flag = args.validate
model_load_path = args.checkpoint_path
############################################
try:
if Net == "deeplab":
net = DeepLab(num_classes = n_classes)
elif Net == "resunet":
net = ResUNet(3, n_classes)
net.apply(weights_init)
elif Net == "unet":
net = UNet(n_channels = 3, n_classes = n_classes)
elif Net == "R2U_Net":
net = R2U_Net(img_ch=3,output_ch=n_classes,t=3)
elif Net == "AttU_Net":
net = AttU_Net(img_ch=3,output_ch=n_classes)
elif Net == "R2AttU_Net":
net = R2AttU_Net(img_ch=3,output_ch=n_classes,t=3)
elif Net == "U_Net":
net = U_Net(img_ch=3,output_ch=n_classes)
net.to(device = device)
if multi_gpu > 1:
net = nn.parallel.DataParallel(net)
############################################
optimizer = optim.Adam(net.parameters())
if n_classes > 1:
criterion = nn.CrossEntropyLoss()
# criterion = MultiClassDiceLoss()
else:
criterion = nn.BCEWithLogitsLoss()
############################################
# todo transformer
transform_train = transforms.Compose([
#transforms.Resize((307, 409)),
transforms.ToTensor(),
])
transform_val = transforms.Compose([
#transforms.Resize((307, 409)),
transforms.ToTensor(),
])
############################################
n_folds = 4
imageDir = os.path.join(image_root, "digestpath_img_patch")
#maskDir = os.path.join(image_root, "digestpath_mask_patch")
typeNames = ["normal", "low level", "high level"]
#typeNames = [ "high level"]
trainImagePaths = []
validImagePaths = []
testImagePaths = []
for i in range(len(typeNames)):
print(typeNames[i])
subDir = os.path.join(imageDir, typeNames[i])
subjectIds = os.listdir(subDir)
tmpIndex1 = len(subjectIds) // n_folds
tmpIndex2 = len(subjectIds) // n_folds * 2
for subjectId in subjectIds[tmpIndex2:]:
subjectDir = os.path.join(subDir, subjectId)
for fileName in os.listdir(subjectDir):
filePath = os.path.join(subjectDir, fileName)
trainImagePaths.append(filePath)
for subjectId in subjectIds[:tmpIndex1]:
subjectDir = os.path.join(subDir, subjectId)
for fileName in os.listdir(subjectDir):
filePath = os.path.join(subjectDir, fileName)
validImagePaths.append(filePath)
for subjectId in subjectIds[tmpIndex1:tmpIndex2]:
#for subjectId in subjectIds:
subjectDir = os.path.join(subDir, subjectId)
for fileName in os.listdir(subjectDir):
filePath = os.path.join(subjectDir, fileName)
testImagePaths.append(filePath)
trainMaskPaths = [p.replace("/digestpath_img_patch/", "/digestpath_mask_patch/")[:-4]+".npy" for p in trainImagePaths]
validMaskPaths = [p.replace("/digestpath_img_patch/", "/digestpath_mask_patch/")[:-4]+".npy" for p in validImagePaths]
testMaskPaths = [p.replace("/digestpath_img_patch/", "/digestpath_mask_patch/")[:-4]+".npy" for p in testImagePaths]
print("train-set #", len(trainMaskPaths))
print("valid-set #", len(validMaskPaths))
print("test-set #", len(testMaskPaths))
############################################
train_set = wsiDataset(trainImagePaths, trainMaskPaths, transform_train)
train_loader = DataLoader(train_set, batch_size=batch_size_train, num_workers=4, shuffle=True)
val_set = wsiDataset(validImagePaths, validMaskPaths, transform_val)
val_loader = DataLoader(val_set, batch_size=1, num_workers=4, shuffle=False)
test_set = wsiDataset(testImagePaths, testMaskPaths, transform_val)
test_loader = DataLoader(test_set, batch_size=1, num_workers=4, shuffle=False)
############################################
if validate_flag:
if os.path.exists(model_load_path):
net.load_state_dict(torch.load(model_load_path, map_location = device))
logging.info(f'Checkpoint loaded from {model_load_path}')
validate_args = {"n_classes":n_classes, "checkpoint_dir":checkpoint_dir, "checkpoint_space":checkpoint_space,}
diceRes, iouRes = validate(test_loader, net, criterion, device, validate_args)
logging.info(f'Test-dataset dice: {diceRes} !')
logging.info(f'Test-dataset iou: {iouRes} !')
else:
logging.info(f'No such checkpoint !')
else:
# todo lr_scheduler
train_args = {"n_classes":n_classes, "epochs":epochs, "checkpoint_dir":checkpoint_dir, "checkpoint_space":checkpoint_space,}
train(train_loader, val_loader, net, criterion, optimizer, device, train_args)
except KeyboardInterrupt:
torch.save(net.state_dict(), os.path.join(checkpoint_dir,'INTERRUPTED.pth'))
logging.info('Saved interrupt')
try:
sys.exit(0)
except SystemExit:
os._exit(0)
def __init__(self, root=r"C:\datasets\taco\tensors", augment_prob=0.0, reduce=0.0, image_size=84,
tensors=True,
random_seed=42, **kwargs):
self.reduce = reduce
self.tensors = tensors
random.seed(random_seed)
resize_train = transforms.Compose(
[
transforms.Resize(image_size),
transforms.CenterCrop(image_size),
]
)
resize_test = transforms.Compose(
[
transforms.Resize(image_size),
transforms.CenterCrop(image_size),
]
)
augment = transforms.Compose(
[
# transforms.RandomRotation(degrees=15),
transforms.RandomHorizontalFlip(p=0.5),
# transforms.ColorJitter(),
# transforms.RandomPerspective(p=0.2, distortion_scale=0.25),
]
)
normalize = transforms.Compose(
[
transforms.ToTensor(),
transforms.Normalize(mean=MEAN, std=STD)
]
)
self.test_transform = transforms.Compose(
[
resize_test,
normalize
]
)
self.train_transform = transforms.Compose(
[
resize_train,
transforms.RandomApply([augment], p=augment_prob),
normalize
]
)
if not tensors:
self.source_dataset_train = torchvision.datasets.ImageFolder(root=root)
else:
self.source_dataset_train = torchvision.datasets.DatasetFolder(root=root, loader=tensor_loader,
extensions=('pt',))
self.dataset_train_size = len(self.source_dataset_train)
items = []
labels = []
for i in range(self.dataset_train_size):
items.append(ImageItem(self.source_dataset_train, i))
labels.append(self.source_dataset_train[i][1])
is_test = [0] * self.dataset_train_size
super(TacoDataset, self).__init__(items, labels, is_test)
self.train_subdataset, self.test_subdataset = self.subdataset.train_test_split()
if reduce < 1:
self.train_subdataset = self.train_subdataset.downscale(1 - reduce)
else:
self.train_subdataset = self.train_subdataset.balance(reduce)
query_dict[query_part] = np.random.choice([query_image for query_image in glob(os.path.join(query_folder, '%03d_*.png') %int(query_part))], arg.num_te)
else:
query_dict[query_part] = [query_image for query_image in glob(os.path.join(query_folder, '%03d_*.png') %int(query_part))]
else:
query_part_set = set([part_image.split('_')[0] for part_image in os.listdir(query_folder)])
query_dict = dict()
for query_part in query_part_set:
query_dict[query_part] = [query_image for query_image in glob(os.path.join(cad_base, furniture, query_part, '*.png'))]
## Input surface vs Query surface (Similarity Comparison)
# Transformation
transform_input = transforms.Compose([
transform_.Padded(),
transform_.RandomResize(max_size=512, ratio=1.5),
transforms.Resize([512, 512]),
transform_.Binary(criterion=150),
transforms.ToTensor(),
transform_.MintoMax()
])
transform_query = transforms.Compose([
transforms.Resize([512, 512]),
transform_.Binary(criterion=150),
transforms.ToTensor(),
transform_.MintoMax()
])
transform3D = transforms.Compose([
transforms.Resize([512, 512]),
transforms.ToTensor(),])
import os
import glob
import cv2
import numpy as np
import torch
from torchvision.transforms import transforms
from natsort import natsorted
from models import resmasking_dropout1
from utils.datasets.fer2013dataset import EMOTION_DICT
from barez import show
transform = transforms.Compose(
[
transforms.ToPILImage(),
transforms.ToTensor(),
]
)
def activations_mask(tensor):
tensor = torch.squeeze(tensor, 0)
tensor = torch.mean(tensor, 0)
tensor = tensor.detach().cpu().numpy()
tensor = np.maximum(tensor, 0)
tensor = cv2.resize(tensor, (224, 224))
tensor = tensor - np.min(tensor)
tensor = tensor / np.max(tensor)
heatmap = cv2.applyColorMap(np.uint8(255 * tensor), cv2.COLORMAP_JET)
return heatmap
# Title: 测试Dsb2018CellDataset数据集.
# URL:
# Date: 2020-12-23
#
import torch
from torch.utils.data import DataLoader
from torchvision.transforms import transforms
from unetzoo.dataset.dataset import DSB2018CellDataset
if __name__ == '__main__':
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
x_transforms = transforms.Compose([
transforms.ToTensor(), # -> [0,1]
transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]) # ->[-1,1]
])
# mask只需要转换为tensor
y_transforms = transforms.ToTensor()
train_dataset = DSB2018CellDataset(r"train", transform=x_transforms, target_transform=y_transforms)
train_dataloaders = DataLoader(train_dataset, batch_size=8)
val_dataset = DSB2018CellDataset(r"val", transform=x_transforms, target_transform=y_transforms)
val_dataloaders = DataLoader(val_dataset, batch_size=1)
test_dataloaders = val_dataloaders
print(len(train_dataloaders.dataset))
for x, y, _, mask in train_dataloaders:
inputs = x
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
init_seeds(0)
model = SVHN_Model1().cuda()
test_path = sorted(glob.glob('D:/Projects/wordec/input/test/*.png'))
# test_json = json.load(open('../input/test.json'))
test_label = [[1]] * len(test_path)
# print(len(test_path), len(test_label))
test_loader = torch.utils.data.DataLoader(
SVHNDataset(test_path, test_label,
transforms.Compose([
transforms.Resize((70, 140)),
# transforms.RandomCrop((60, 120)),
# transforms.ColorJitter(0.3, 0.3, 0.2),
# transforms.RandomRotation(5),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [
0.229, 0.224, 0.225])
])),
batch_size=40,
shuffle=False,
num_workers=0,
)
# 加载保存的最优模型
model.load_state_dict(torch.load('D:/Projects/wordec/model.pt'))
test_predict_label = predict(test_loader, model, 1)
print(test_predict_label.shape)
print('test_predict_label', test_predict_label)
def main():
args = parse_args()
FRAME_LIMIT = args.frame_limit
VIDEO_FILE = args.video_file
DRAW_BOXES = args.draw_boxes
SCORES_FILE = args.scores_file
FOLDER = args.output_folder
OUTPUT_VIDEO = args.output_video
FACE_DETECTOR = args.face_detector
USE_WEBCAM = args.use_webcam
TRACK_BOXES = args.track_boxes
THRESHOLD = args.threshold
# load configs and set random seed
configs = json.load(open("./configs/fer2013_config.json"))
image_size = (configs["image_size"], configs["image_size"])
transform = transforms.Compose([transforms.ToPILImage(), transforms.ToTensor()])
detector = load_face_detector(FACE_DETECTOR)
model = load_model()
vid = cv2.VideoCapture(0)
with torch.no_grad():
if USE_WEBCAM:
while True:
ret, img = vid.read()
if img is None or ret is not True:
continue
faces = detector.detect_faces(img, THRESHOLD)
img, _ = detect_emotions(model, img, faces, image_size, transform, DRAW_BOXES, TRACK_BOXES)
cv2.imshow("img", img)
if cv2.waitKey(1) == ord("q"):
break
cv2.destroyAllWindows()
else:
print(f">> Reading video from {VIDEO_FILE}")
frames = collect_frames(VIDEO_FILE, FRAME_LIMIT)
if TRACK_BOXES:
df = pd.read_csv(SCORES_FILE, header=None, index_col=None, names=[
"track_id", "frame_id", "box_id", "x", "y", "x2", "y2"] + list(range(80)))
df["abs_frame_id"] = df.frame_id + df.track_id - 128
df_faces = []
print(">> Detecting emotions")
for i, frame in tqdm(enumerate(frames), total=len(frames)):
faces = match_faces_bodies(
frame,
df[df.abs_frame_id == i],
detector=detector,
threshold=THRESHOLD
)
frames[i], new_df_faces = detect_emotions(model, frame, faces, image_size, transform, DRAW_BOXES, TRACK_BOXES)
df_faces += new_df_faces
df_faces = pd.DataFrame(df_faces)
print(f">> Saving box files at {FOLDER}")
if os.path.exists(FOLDER):
shutil.rmtree(FOLDER)
os.mkdir(FOLDER)
box_ids = df_faces.box_id.unique()
for box_id in tqdm(box_ids, total=len(box_ids)):
with open(os.path.join(FOLDER, f"person{box_id}.txt"), "w") as f:
for _, frame in df_faces[df_faces.box_id == box_id].iterrows():
bbox = str([frame.x, frame.y, frame.x2, frame.y2]).replace(" ", "").strip("[]")
emotion = list(frame.emotion.keys())[0]
f.write(f"{frame.frame_id},{bbox},{emotion},{frame.emotion[emotion]}\n")
else:
print(">> Detecting emotions")
for i, frame in tqdm(enumerate(frames), total=len(frames)):
faces = detector.detect_faces(frame, THRESHOLD)
frames[i], _ = detect_emotions(model, frame, faces, image_size, transform, DRAW_BOXES, TRACK_BOXES)
print(f">> Creating video at {OUTPUT_VIDEO}")
height, width, _ = frames[0].shape
out = cv2.VideoWriter(OUTPUT_VIDEO, cv2.VideoWriter_fourcc(*"mp4v"), 25, (width, height), True)
for frame in tqdm(frames, total=len(frames)):
out.write(frame)
cv2.destroyAllWindows()
out.release()
"xresnet18": xresnet18ssa,
"xresnet34": xresnet34ssa
}
Model = model_dict[args.model]
train_df = pd.read_csv(args.train_df_path)
processed = pickle.load(open(args.train_processed_path, "rb"))
test_fns = sorted(os.listdir(args.test_dir))
test_df = pd.DataFrame()
test_df["File"] = test_fns
processed_test = pickle.load(open(args.test_processed_path, "rb"))
transforms_dict = {
'train':
transforms.Compose([
transforms.RandomHorizontalFlip(0.5),
transforms.ToTensor(),
]),
'test':
transforms.Compose([
transforms.RandomHorizontalFlip(0.5),
transforms.ToTensor(),
]),
}
x_train = processed
y_train = train_df.Label.values.astype(np.long)
x_test = processed_test
splits = list(
StratifiedKFold(n_splits=5, shuffle=True,
random_state=SEED).split(x_train, y_train))
def get_transforms(config, image_size=None):
config = config.get_dictionary()
if image_size is not None:
image_size = image_size
elif config['estimator'] not in resize_size_dict:
image_size = 32
else:
image_size = resize_size_dict[config['estimator']]
val_transforms = transforms.Compose([
transforms.Resize(image_size),
transforms.CenterCrop(image_size),
transforms.ToTensor(),
])
if parse_bool(config['aug']):
if parse_bool(config['auto_aug']):
# from .transforms import AutoAugment
data_transforms = {
'train':
transforms.Compose([
# AutoAugment(),
transforms.Resize(image_size),
transforms.RandomCrop(image_size,
padding=int(image_size / 8)),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
]),
'val':
val_transforms,
}
else:
transform_list = []
if parse_bool(config['jitter']):
transform_list.append(
transforms.ColorJitter(brightness=config['brightness'],
saturation=config['saturation'],
hue=config['hue']))
if parse_bool(config['affine']):
transform_list.append(
transforms.RandomAffine(degrees=config['degree'],
shear=config['shear']))
transform_list.append(transforms.RandomResizedCrop(image_size))
transform_list.append(transforms.RandomCrop(image_size, padding=4))
if parse_bool(config['random_flip']):
transform_list.append(transforms.RandomHorizontalFlip())
transform_list.append(transforms.ToTensor())
data_transforms = {
'train': transforms.Compose(transform_list),
'val': val_transforms
}
else:
data_transforms = {
'train':
transforms.Compose([
transforms.Resize(image_size),
transforms.CenterCrop(image_size),
transforms.ToTensor(),
]),
'val':
val_transforms,
}
return data_transforms
def __init__(
self,
c,
nof_joints,
checkpoint_path,
model_name='HRNet',
resolution=(384, 288),
interpolation=cv2.INTER_CUBIC,
multiperson=True,
return_heatmaps=False,
return_bounding_boxes=False,
max_batch_size=32,
yolo_model_def="./models/detectors/yolo/config/yolov3.cfg",
yolo_class_path="./models/detectors/yolo/data/coco.names",
yolo_weights_path="./models/detectors/yolo/weights/yolov3.weights",
device=torch.device("cpu")):
"""
Initializes a new SimpleHRNet object.
HRNet (and YOLOv3) are initialized on the torch.device("device") and
its (their) pre-trained weights will be loaded from disk.
Args:
c (int): number of channels (when using HRNet model) or resnet size (when using PoseResNet model).
nof_joints (int): number of joints.
checkpoint_path (str): path to an official hrnet checkpoint or a checkpoint obtained with `train_coco.py`.
model_name (str): model name (HRNet or PoseResNet).
Valid names for HRNet are: `HRNet`, `hrnet`
Valid names for PoseResNet are: `PoseResNet`, `poseresnet`, `ResNet`, `resnet`
Default: "HRNet"
resolution (tuple): hrnet input resolution - format: (height, width).
Default: (384, 288)
interpolation (int): opencv interpolation algorithm.
Default: cv2.INTER_CUBIC
multiperson (bool): if True, multiperson detection will be enabled.
This requires the use of a people detector (like YOLOv3).
Default: True
return_heatmaps (bool): if True, heatmaps will be returned along with poses by self.predict.
Default: False
return_bounding_boxes (bool): if True, bounding boxes will be returned along with poses by self.predict.
Default: False
max_batch_size (int): maximum batch size used in hrnet inference.
Useless without multiperson=True.
Default: 16
yolo_model_def (str): path to yolo model definition file.
Default: "./models/detectors/yolo/config/yolov3.cfg"
yolo_class_path (str): path to yolo class definition file.
Default: "./models/detectors/yolo/data/coco.names"
yolo_weights_path (str): path to yolo pretrained weights file.
Default: "./models/detectors/yolo/weights/yolov3.weights.cfg"
device (:class:`torch.device`): the hrnet (and yolo) inference will be run on this device.
Default: torch.device("cpu")
"""
self.c = c
self.nof_joints = nof_joints
self.checkpoint_path = checkpoint_path
self.model_name = model_name
self.resolution = resolution # in the form (height, width) as in the original implementation
self.interpolation = interpolation
self.multiperson = multiperson
self.return_heatmaps = return_heatmaps
self.return_bounding_boxes = return_bounding_boxes
self.max_batch_size = max_batch_size
self.yolo_model_def = yolo_model_def
self.yolo_class_path = yolo_class_path
self.yolo_weights_path = yolo_weights_path
self.device = device
if model_name in ('HRNet', 'hrnet'):
self.model = HRNet(c=c, nof_joints=nof_joints)
elif model_name in ('PoseResNet', 'poseresnet', 'ResNet', 'resnet'):
self.model = PoseResNet(resnet_size=c, nof_joints=nof_joints)
else:
raise ValueError('Wrong model name.')
checkpoint = torch.load(checkpoint_path, map_location=self.device)
if 'model' in checkpoint:
self.model.load_state_dict(checkpoint['model'])
else:
self.model.load_state_dict(checkpoint)
if 'cuda' in str(self.device):
print("device: 'cuda' - ", end="")
if 'cuda' == str(self.device):
# if device is set to 'cuda', all available GPUs will be used
print("%d GPU(s) will be used" % torch.cuda.device_count())
device_ids = None
else:
# if device is set to 'cuda:IDS', only that/those device(s) will be used
print("GPU(s) '%s' will be used" % str(self.device))
device_ids = [int(x) for x in str(self.device)[5:].split(',')]
self.model = torch.nn.DataParallel(self.model,
device_ids=device_ids)
elif 'cpu' == str(self.device):
print("device: 'cpu'")
else:
raise ValueError('Wrong device name.')
self.model = self.model.to(device)
self.model.eval()
if not self.multiperson:
self.transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
])
else:
self.detector = YOLOv3(model_def=yolo_model_def,
class_path=yolo_class_path,
weights_path=yolo_weights_path,
classes=('person', ),
max_batch_size=self.max_batch_size,
device=device)
self.transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize((self.resolution[0],
self.resolution[1])), # (height, width)
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
])
x = self.layerblock3(x)
x = self.layerblock4(x)
x = self.layerblock5(x)
x = x.view(x.size(0), -1)
x = self.classifier(x)
return x
if __name__ == "__main__":
filepath = './CIFAR10_dataset'
#data loading and augmentation
#cifar 10 normalization parameter https://arxiv.org/pdf/1909.12205.pdf
train_transformations = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.RandomCrop(32, padding=4),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.247, 0.243, 0.261))
])
test_transformations = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.247, 0.243, 0.261))
])
train_dataset = CIFAR10(root=filepath,
train=True,
download=True,
transform=train_transformations)
test_dataset = CIFAR10(root=filepath,
train=False,
download=True,
transform=test_transformations)
batch_size = 100
plt.subplot(sqrtn, sqrtn, index+1)
plt.imshow(image.reshape(28, 28))
device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
print('GPU State:', device)
G = generator().to(device)
D = discriminator().to(device)
print(G)
print(D)
epochs = 200
lr = 0.0002
batch_size = 64
g_optimizer = optim.Adam(G.parameters(), lr=lr, betas=(0.5, 0.999))
d_optimize = optim.Adam(D.parameters(), lr=lr, betas=(0.5, 0.999))
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,))])
train_set = datasets.MNIST('mnist/', train=True, download=False, transform=transform)
test_set = datasets.MNIST('mnist/', train=False, download=False, transform=transform)
train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True)
test_loader = DataLoader(test_set, batch_size=batch_size, shuffle=False)
# Train
for epoch in range(epochs):
epoch += 1
for time, data in enumerate(train_loader):
time += 1
true_input = data[0].to(device)
test = 255 * (0.5 * true_input[0] + 0.5)
from pfrl import replay_buffers, explorers, experiments, q_functions
from pfrl.agents import DQN
from pfrl.q_functions import DiscreteActionValueHead
from torch import nn, optim
from torchvision.datasets import CIFAR10
from torchvision.transforms import transforms
import numpy as np
from tqdm import tqdm, trange
from rl.env_d import AttackEnv
if __name__ == '__main__':
transform = transforms.Compose((transforms.ToTensor(), transforms.Normalize(0.5, 0.5, 0.5)))
test_dataset = CIFAR10('data', train=False, transform=transform, download=False)
image_size = 32 * 32
n_classes = 10
max_episodes = 20
max_episode_len = 1000
env = AttackEnv()
obs_space = env.observation_space
action_space = env.action_space
obs_size = obs_space.low.size
n_actions = action_space.n
('arch', 'lenet5'),
('n_classes', 10),
('input_shape', (1, 28, 28)),
])
data_config = OrderedDict([
('dataset', 'SplitFashionMNIST'),
('valid', 0.0),
('num_workers', 4),
(
'train_transform',
transforms.Compose([
lambda x: np.array(x).reshape((1, 28, 28)),
lambda x: np.pad(x, (
(0, 0), (2, 2),
(2, 2)), mode='minimum'), # Padding is only required by LeNet
lambda x: torch.FloatTensor(x),
lambda x: x / 255.0,
transforms.Normalize(np.array([0.1307]), np.array([0.3081]))
])),
(
'test_transform',
transforms.Compose([
lambda x: np.array(x).reshape((1, 28, 28)),
lambda x: np.pad(x, (
(0, 0), (2, 2),
(2, 2)), mode='minimum'), # Padding is only required by LeNet
lambda x: torch.FloatTensor(x),
lambda x: x / 255.0,
transforms.Normalize(np.array([0.1307]), np.array([0.3081]))
])),
format_string = self.__class__.__name__ + '(size={0}'.format(self.size)
format_string += ', xscale={0}'.format(
tuple(round(s, 4) for s in self.xscale))
format_string += ', yscale={0}'.format(
tuple(round(r, 4) for r in self.xscale))
format_string += ', interpolation={0})'.format(interpolate_str)
return format_string
transform = {
'train':
transforms.Compose([
transforms.RandomRotation(12, resample=Image.BILINEAR),
CenterRandomCrop(size, xscale=(0.6, 1.0), aspect_ratio=(1, 1.4)),
transforms.ColorJitter(0.2, 0.1, 0.1, 0.04),
transforms.RandomHorizontalFlip(),
transforms.RandomVerticalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean, std)
]),
'val':
transforms.Compose([
transforms.CenterCrop((512, 640)),
transforms.Resize(size, interpolation=Image.BILINEAR),
transforms.ToTensor(),
transforms.Normalize(mean, std)
])
}
if torch.cuda.is_available():
device = torch.device("cuda:0")
from data_loader.DataPath import DataPath
from framework.loss import Loss
from config import ParallelConfig
from framework.segmentation.Mask import Mask
from segmentation.loss.neighbour_diff import NeighbourDiffLoss
from framework.segmentation.unet import UNetSegmentation
from viz.visualization import show_segmentation, show_images
LABELS_URL = 'https://s3.amazonaws.com/mlpipes/pytorch-quick-start/labels.json'
labels = {int(key):value for (key, value)
in requests.get(LABELS_URL).json().items()}
transform = transforms.Compose([ #[1]
transforms.Resize(128), #[2]
transforms.CenterCrop(128), #[3]
transforms.ToTensor(), #[4]
transforms.Normalize( #[5]
mean=[0.485, 0.456, 0.406], #[6]
std=[0.229, 0.224, 0.225] #[7]
)])
mask_transform = transforms.Compose([ #[1]
transforms.Resize(128), #[2]
transforms.CenterCrop(128), #[3]
transforms.ToTensor() #[4]
])
# trainset = CocoStuff10k(root="/home/nazar/PycharmProjects/coco", transform=transform)
trainset = Cityscapes(DataPath.HOME_STREET, transform=transform, target_transform=mask_transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=20, shuffle=True, num_workers=12)
F.cuda()
# load saved models
G.load_state_dict(torch.load('../save/G.pth'))
F.load_state_dict(torch.load('../save/F.pth'))
# set to eval states
G.eval()
F.eval()
Tensor = torch.cuda.FloatTensor if cuda else torch.Tensor
input_X = Tensor(batch_size, input_nc, image_size, image_size)
input_Y = Tensor(batch_size, output_nc, image_size, image_size)
transforms = transforms.Compose([
transforms.Resize((image_size, image_size)),
transforms.ToTensor(),
])
dataset = ImageDataset(dataroot=dataroot, transforms=transforms, mode='test')
dataloader = DataLoader(dataset=dataset, batch_size=batch_size, shuffle=False, num_workers=num_workers)
output_dir_X = '../output/X'
output_dir_Y = '../output/Y'
output_dir_recover = '../output/recover'
if not os.path.exists(output_dir_X):
os.makedirs(output_dir_X)
if not os.path.exists(output_dir_Y):
os.makedirs(output_dir_Y)
if not os.path.exists(output_dir_recover):
os.makedirs(output_dir_recover)
def gen_transform_test_multi(crop_size=448):
return transforms.Compose([
transforms.RandomCrop(crop_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5], std=[0.5])
])
#save = "/DB/VCDB/all_frames/resnet50-rmac"
if not os.path.exists(base):
print("base '{}' is not exist.".format(base))
exit()
if not os.path.exists(save):
os.makedirs(os.path.join(save, 'v-features'))
os.makedirs(os.path.join(save, 'f-features'))
os.makedirs(os.path.join(save2, 'v-features'))
os.makedirs(os.path.join(save2, 'f-features'))
videos = os.listdir(base)
videos.sort(key=int)
normalize = trn.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
video_trn = trn.Compose([trn.Resize(224), trn.ToTensor(), normalize])
model2 = resnet50(pretrained=True)
model2 = torch.nn.Sequential(*list(model2.children())[:-1]) # 2048 - dimension
model2.cuda()
model2 = torch.nn.DataParallel(model2)
summary(model2, (3, 224, 224))
model2.eval()
model = Resnet50_RMAC()
model.cuda()
model = torch.nn.DataParallel(model)
summary(model, (3, 224, 224))
model.eval()
with torch.no_grad():
def __getitem__(self, index):
'''
over-writing the __getitem__ method to return the index
along with img and target
'''
img, target = super().__getitem__(index)
# print(f"mnist index: {index}")
return img, target, index
if __name__ == "__main__":
transform = transforms.Compose([
# you can add other transformations in this list
transforms.ToTensor()
])
temp_MNIST = CustomMNIST(root='./data',
train=True,
download=True,
transform=transform)
train_split_percentage = 0.2
initial_train_set_size = len(temp_MNIST)
needed_train_set_size = int(train_split_percentage *
initial_train_set_size)
needed_val_set_size = initial_train_set_size - needed_train_set_size
print(needed_train_set_size, needed_val_set_size)
train_set, val_set = torch.utils.data.random_split(
temp_MNIST, (needed_train_set_size, needed_val_set_size))
def encode(img, bottleneck):
transform = transforms.Compose([transforms.ToTensor()])
inputs = transform(img).unsqueeze(0)
encoder = Encoder()
binarizer = Binarizer(int(bottleneck / 512))
decoder = Decoder(int(bottleneck / 512))
if torch.cuda.is_available():
encoder = Encoder().cuda()
binarizer = Binarizer(int(bottleneck / 512)).cuda()
decoder = Decoder(int(bottleneck / 512)).cuda()
encoder.load_state_dict(
torch.load('./project/models/model_LSTM_' + str(bottleneck) +
'/encoder'))
binarizer.load_state_dict(
torch.load('./project/models/model_LSTM_' + str(bottleneck) +
'/binarizer'))
decoder.load_state_dict(
torch.load('./project/models/model_LSTM_' + str(bottleneck) +
'/decoder'))
else:
encoder.load_state_dict(
torch.load('./project/models/model_LSTM_' + str(bottleneck) +
'/encoder',
map_location=torch.device('cpu')))
binarizer.load_state_dict(
torch.load('./project/models/model_LSTM_' + str(bottleneck) +
'/binarizer',
map_location=torch.device('cpu')))
decoder.load_state_dict(
torch.load('./project/models/model_LSTM_' + str(bottleneck) +
'/decoder',
map_location=torch.device('cpu')))
encoder.eval()
binarizer.eval()
decoder.eval()
if torch.cuda.is_available():
e_1 = (torch.zeros(1, 256, 64, 64).cuda(), torch.zeros(1, 256, 64,
64).cuda())
e_2 = (torch.zeros(1, 512, 32, 32).cuda(), torch.zeros(1, 512, 32,
32).cuda())
e_3 = (torch.zeros(1, 512, 16, 16).cuda(), torch.zeros(1, 512, 16,
16).cuda())
d_1 = (torch.zeros(1, 512, 16, 16).cuda(), torch.zeros(1, 512, 16,
16).cuda())
d_2 = (torch.zeros(1, 512, 32, 32).cuda(), torch.zeros(1, 512, 32,
32).cuda())
d_3 = (torch.zeros(1, 256, 64, 64).cuda(), torch.zeros(1, 256, 64,
64).cuda())
d_4 = (torch.zeros(1, 128, 128,
128).cuda(), torch.zeros(1, 128, 128, 128).cuda())
residual = inputs.cuda()
else:
e_1 = (torch.zeros(1, 256, 64, 64), torch.zeros(1, 256, 64, 64))
e_2 = (torch.zeros(1, 512, 32, 32), torch.zeros(1, 512, 32, 32))
e_3 = (torch.zeros(1, 512, 16, 16), torch.zeros(1, 512, 16, 16))
d_1 = (torch.zeros(1, 512, 16, 16), torch.zeros(1, 512, 16, 16))
d_2 = (torch.zeros(1, 512, 32, 32), torch.zeros(1, 512, 32, 32))
d_3 = (torch.zeros(1, 256, 64, 64), torch.zeros(1, 256, 64, 64))
d_4 = (torch.zeros(1, 128, 128, 128), torch.zeros(1, 128, 128, 128))
residual = inputs
binary = []
for t in range(rnn_num):
e_result, e_1, e_2, e_3 = encoder(residual, e_1, e_2, e_3)
b_result = binarizer(e_result)
outputs, d_1, d_2, d_3, d_4 = decoder(b_result, d_1, d_2, d_3, d_4)
residual = residual - outputs
binary.append(
(torch.Tensor.cpu(b_result).detach().numpy().astype(np.int8) + 1)
// 2)
binary = np.stack(binary, axis=0)
binary = np.packbits(binary, axis=-1)
return binary
print("Checking ......")
args = get_args()
check_args(args)
print("Initializaion ......")
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
#prepare data
train_traisform = transforms.Compose([
transforms.RandomResizedCrop(args.img_size),
#transforms.Resize((args.img_size,args.img_size)),
transforms.RandomHorizontalFlip(),
#transforms.ColorJitter(brightness=0.4, contrast=0.4, saturation=0.4),
transforms.ToTensor(),
normalize,
])
if args.crop:
val_traisform = transforms.Compose([
transforms.Resize(args.img_size),
transforms.CenterCrop(args.img_size),
#transforms.ColorJitter(brightness=0.4, contrast=0.4, saturation=0.4),
transforms.ToTensor(),
normalize,
])
else:
val_traisform = transforms.Compose([
transforms.Resize((args.img_size, args.img_size)),
def __init__(
self,
dataset_path='./drive/MyDrive/datasets/car classification/train_data',
batch_size=1,
model_name='tf_efficientnet_b3_ns',
test_csv='./train_labels.csv',
unique_csv='./train_labels.csv',
output_dir='../drive/MyDrive/ckpt/grapheme/submission.csv',
ckpt='../drive/MyDrive/ckpt/grapheme/20.pth'):
# initialize attributes
self.dataset_path = dataset_path
self.batch_size = batch_size
self.model_name = model_name
self.test_csv = test_csv
self.unique_csv = unique_csv
self.output_dir = output_dir
self.ckpt = ckpt
if model_name == 'tf_efficientnet_b0_ns':
self.input_size = (224, 224)
elif model_name == 'tf_efficientnet_b3_ns':
self.input_size = (300, 300)
elif model_name == 'tf_efficientnet_b4_ns':
scaleelf.input_size = (380, 380)
elif model_name == 'tf_efficientnet_b6_ns':
self.input_size = (528, 528)
else:
raise Exception('non-valid model name')
# Compose transforms
transform = []
transform += [transforms.Resize(self.input_size)]
self.transform = transforms.Compose(transform)
self.test_dataset = BengaliDataset(self.test_csv,
self.unique_csv,
self.dataset_path,
self.transform,
cache=True)
self.names = self.test_dataset.names
self.test_dataloader = DataLoader(self.test_dataset,
batch_size=self.batch_size,
num_workers=0,
shuffle=False)
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
self.model_root = MyModel(self.input_size,
self.model_name,
168,
pretrained=True,
dropout=0).to('cuda')
self.model_consonant = MyModel(self.input_size,
self.model_name,
11,
pretrained=True,
dropout=0).to('cuda')
self.model_vowel = MyModel(self.input_size,
self.model_name,
18,
pretrained=True,
dropout=0).to('cuda')
self.model_multihead = MultiHeadModel(self.input_size,
self.model_name,
pretrained=True,
dropout=0).to('cuda')
ckpt = torch.load(self.ckpt)
self.model_root.load_state_dict(ckpt['model_root_state_dict'])
self.model_consonant.load_state_dict(
ckpt['model_consonant_state_dict'])
self.model_vowel.load_state_dict(ckpt['model_vowel_state_dict'])
self.model_multihead.load_state_dict(
ckpt['model_multihead_state_dict'])
