def __init__(self, model_id, project_dir):
super(DeepLab_Plus, self).__init__()
self.model_id = model_id
self.project_dir = project_dir
self.create_model_dirs()
self.network = DeepLabV3(self.model_id, self.project_dir)
self.network.load_state_dict(
torch.load("pretrained_models/model_13_2_2_2_epoch_580.pth",
map_location='cpu'))
self.conv1 = nn.Conv2d(in_channels=512,
out_channels=8,
kernel_size=3,
padding=1)
self.fc = nn.Linear(in_features=1296, out_features=2)
def main():
dataloaders = {
'train':
torch.utils.data.DataLoader(image_datasets['train'],
batch_size=1,
shuffle=True,
num_workers=8),
'val':
torch.utils.data.DataLoader(image_datasets['val'],
batch_size=1,
shuffle=True,
num_workers=8)
}
device = torch.device("cuda:" + str(gpu))
#################################
# labels = [0, 1]
nclasses = len(labels)
# model = Unet2D(1, nclasses)
model = DeepLabV3(len(labels))
model = model.to(device)
# model.half() # convert to half precision
# for layer in model.modules():
# if isinstance(layer, nn.BatchNorm2d):
# layer.float()
#################################
# weights = None
criterion = DiceLoss()
# criterion = nn.CrossEntropyLoss()
optimizer_conv = optim.Adam(model.parameters(), lr=0.01)
# optimizer_conv = Adam16(model.parameters(), lr=1e-3)
exp_lr_scheduler = lr_scheduler.ReduceLROnPlateau(optimizer_conv)
# print(model)
epochs = 20
train_model(model,
criterion,
optimizer_conv,
exp_lr_scheduler,
dataloaders,
device,
num_epochs=epochs)
def segmentation(filename, output_file="output.jpg"):
"""
Input:
filename: location of (1024x2048x3) png input image
output: filename for segmented output image, size (1024x2048x1) with
labels 0-19 for each pixel
Outputs:
output file saved as given name
"""
# Load pre-trained network
model = DeepLabV3(1, ".")
if torch.cuda.is_available():
model.load_state_dict(torch.load("model_13_2_2_2_epoch_580.pth"))
else:
model.load_state_dict(
torch.load("model_13_2_2_2_epoch_580.pth",
map_location=lambda storage, loc: storage))
model.eval()
# Preprocess input
transformation = transforms.Compose([
transforms.ToTensor(),
])
#image = Image.open("stuttgart_00_000000_000080_leftImg8bit.png")
image = Image.open(filename)
image_tensor = transformation(image).float()
image_tensor = image_tensor.unsqueeze_(0)
if torch.cuda.is_available():
image_tensor.cuda()
# Run model
input = Variable(image_tensor)
output = model(input)
# Output
out_img = np.argmax(output.detach().numpy(),
axis=1).astype(np.uint8)[0, :, :]
np.save("output.npy", out_img)
imsave(output_file, out_img)
parser = argparse.ArgumentParser(description='Compute Confusion Matrix')
print(os.getcwd())
parser.add_argument('--model_path',
type=str,
default=os.getcwd() +
'/pretrained_models/model_13_2_2_2_epoch_580.pth')
parser.add_argument('--fog_scale', type=float, default=0.005)
parser.add_argument('--dataset_path', type=str, default='data/cityscapes')
parser.add_argument('--compute_unperturbed', action='store_true')
args = parser.parse_args()
device = 'cuda' if torch.cuda.is_available() else 'cpu'
model_name = '0'
batch_size = 8
model = DeepLabV3(model_name, project_dir=os.getcwd()).to(device)
model.load_state_dict(torch.load(args.model_path))
print(model)
test_datasets1 = DATASET_CITYSCAPES(cityscapes_path='data/cityscapes',
split='val')
test_datasets2 = DATASET_CITYSCAPES_FOGGY(cityscapes_path='data/cityscapes',
split='val',
fog_scale=args.fog_scale)
test_loader1 = torch.utils.data.DataLoader(dataset=test_datasets1,
batch_size=batch_size,
shuffle=False,
num_workers=4)
test_loader2 = torch.utils.data.DataLoader(dataset=test_datasets2,
batch_size=batch_size,
shuffle=False,
num_workers=4)
import torch.utils.data
import torch.nn as nn
from torch.autograd import Variable
import torch.optim as optim
import torch.nn.functional as F
import numpy as np
import pickle
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import cv2
batch_size = 2
network = DeepLabV3("eval_val", project_dir="..").cuda()
network.load_state_dict(
torch.load("../pretrained_models/model_13_2_2_2_epoch_580.pth"))
val_dataset = DatasetVal(cityscapes_data_path="../data/cityscapes",
cityscapes_meta_path="../data/cityscapes/meta")
num_val_batches = int(len(val_dataset) / batch_size)
print("num_val_batches:", num_val_batches)
val_loader = torch.utils.data.DataLoader(dataset=val_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=1)
with open("../data/cityscapes/meta/class_weights.pkl",
def main():
# Training settings
parser = argparse.ArgumentParser(
description='Scratch segmentation Example')
parser.add_argument('--batch-size',
type=int,
default=1,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size',
type=int,
default=1,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=30,
metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr',
type=float,
default=0.001,
metavar='LR',
help='learning rate (default: 0.01)')
parser.add_argument('--momentum',
type=float,
default=0.5,
metavar='M',
help='SGD momentum (default: 0.5)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
parser.add_argument(
'--log-interval',
type=int,
default=10,
metavar='N',
help='how many batches to wait before logging training status')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
print('my device is :', device)
kwargs = {'num_workers': 0, 'pin_memory': True} if use_cuda else {}
train_loader = torch.utils.data.DataLoader(FarmDataset(istrain=True),
batch_size=args.batch_size,
shuffle=True,
drop_last=True,
**kwargs)
startepoch = 0
# model = torch.load('./tmp/model{}'.format(startepoch)) if startepoch else segnet(3, 4).to(device)
# model = torch.load('./tmp/model{}'.format(startepoch)) if startepoch else unet(3, 4).to(device)
# model = torch.load('./tmp/model{}'.format(startepoch)) if startepoch else nested_unet(3, 4).to(device)
model = torch.load('./tmp/model{}'.format(
startepoch)) if startepoch else DeepLabV3().to(device)
args.epochs = 5
# optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum)
learning_rate = 1e-4
optimizer = optim.Adam(model.parameters(), lr=learning_rate)
train_losses, train_mious = [], []
val_losses, val_mious = [], []
for epoch in range(startepoch, args.epochs + 1):
#relu+bn
train_loss, train_miou = train(args, model, device, train_loader,
optimizer, epoch)
val_loss, val_miou = val(args,
model,
device,
valDataset(istrain=True, isaug=False),
issave=False)
train_losses.append(train_loss)
train_mious.append(train_miou)
val_losses.append(val_loss)
val_mious.append(val_miou)
if epoch % 3 == 0:
print(epoch)
val(args,
model,
device,
valDataset(istrain=True, isaug=False),
issave=False)
torch.save(model, './tmp/model{}'.format(epoch))
plt.plot(range(1,
len(train_losses) + 1),
train_losses,
'bo',
label='train loss')
plt.plot(range(1, len(val_losses) + 1), val_losses, 'r', label='val loss')
plt.legend()
plt.show()
plt.plot(range(1,
len(train_mious) + 1),
train_mious,
'bo',
label='train miou')
plt.plot(range(1, len(val_mious) + 1), val_mious, 'r', label='val miou')
plt.legend()
plt.show()
rows, cols = 5, 2 #Show 10 images in the dataset along with target and predicted masks
device = torch.device("cuda")# if torch.cuda.is_available() else "cpu")
num_gpu = list(range(torch.cuda.device_count()))
testloader, test_dst = load_dataset(batch_size, num_workers, split=test_split)
# Creating an instance of the model
#model = Segnet(n_classes) #Fully Convolutional Networks
#model = U_Net(img_ch=3,output_ch=n_classes) #U Network
#model = R2U_Net(img_ch=3,output_ch=n_classes,t=2) #Residual Recurrent U Network, R2Unet (t=2)
#model = R2U_Net(img_ch=3,output_ch=n_classes,t=3) #Residual Recurrent U Network, R2Unet (t=3)
#model = RecU_Net(img_ch=3,output_ch=n_classes,t=2) #Recurrent U Network, RecUnet (t=2)
#model = ResU_Net(img_ch=3,output_ch=n_classes) #Residual U Network, ResUnet
#model = DeepLabV3(n_classes, 'vgg') #DeepLabV3 VGG backbone
model = DeepLabV3(n_classes, 'resnet') #DeepLabV3 Resnet backbone
print('Evaluation logs for model: {}'.format(model.__class__.__name__))
model = nn.DataParallel(model, device_ids=num_gpu).to(device)
model_params = torch.load(os.path.join(expt_logdir, "{}".format(ckpt_name)))
model.load_state_dict(model_params)
#Visualization of test data
test_vis = Vis(test_dst, expt_logdir, rows, cols)
#Metrics calculator for test data
test_metrics = Metrics(n_classes, testloader, test_split, device, expt_logdir)
model.eval()
for i, (inputs, labels) in enumerate(testloader):
inputs = inputs.to(device)
import matplotlib.pyplot as plt
import cv2
import glob
def getEpoch(checkpoint_name):
filename_w_ext = os.path.basename(checkpoint_name)
filename, file_extension = os.path.splitext(filename_w_ext)
filenames = filename.split("_")
return filenames[3]
batch_size = 2
model_id = 1
network = DeepLabV3("eval_val", project_dir=default_path).cuda()
#network.load_state_dict(torch.load(os.path.join(default_path,'training_logs/model_1/checkpoints/model_1_epoch_251.pth')))
#check last checkpoint
data_list = glob.glob(
os.path.join(
default_path, 'training_logs/model_' + str(model_id) +
'/checkpoints/model_' + str(model_id) + '_*.pth'))
#find latest checkpoint
latest_epoch = 0
for name in list(data_list):
if latest_epoch < int(getEpoch(name)):
latest_epoch = int(getEpoch(name))
if latest_epoch != 0:
network.load_state_dict(
torch.load(
train_data = VOCSegmentation('data/')
val_data = VOCSegmentation(
'data/',
image_set='val',
)
elif dataset_type == 'cityscapes':
train_data = Cityscapes('data/', h=256, w=512)
val_data = Cityscapes('data/', split='val', h=256, w=512)
elif dataset_type == 'fss1000':
train_data = FSS1000('data/', h=256, w=256)
val_data = FSS1000('data', image_set='val', h=256, w=256)
train_loader = DataLoader(train_data, batch_size=BATCH_SIZE, shuffle=True)
val_loader = DataLoader(val_data, batch_size=BATCH_SIZE, shuffle=False)
model = DeepLabV3(num_classes=NUM_CLASSES)
if resume_training:
if dataset_type == 'voc':
model = torch.load(f'{checkpoint_dir}/voc.pt')
elif dataset_type == 'cityscapes':
model = torch.load(f'{checkpoint_dir}/cityscapes.pt')
elif dataset_type == 'fss1000':
model = torch.load(f'{checkpoint_dir}/fss1000.pt')
model.to(device)
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(),
lr=LEARNING_RATE,
weight_decay=WEIGHT_DECAY)
10: 23,
11: 24,
12: 25,
13: 26,
14: 27,
15: 28,
16: 31,
17: 32,
18: 33,
19: 0
}
trainId_to_id_map_func = np.vectorize(trainId_to_id.get)
batch_size = 2
network = DeepLabV3("val_result", project_dir="..").cuda()
network.load_state_dict(
torch.load("../pretrained_models/model_13_2_2_2_epoch_580.pth"))
val_dataset = DatasetVal(cityscapes_data_path="../data/cityscapes",
cityscapes_meta_path="../data/cityscapes/meta")
num_val_batches = int(len(val_dataset) / batch_size)
print("num_val_batches:", num_val_batches)
val_loader = torch.utils.data.DataLoader(dataset=val_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=1)
with open("../data/cityscapes/meta/class_weights.pkl",
import numpy as np
from dataset import CityScapeDataSet
from deeplabv3 import DeepLabV3
from score import SegmentationMetric
# dataset
batch_size = 2
dataset = CityScapeDataSet()
data_loader = data.DataLoader(dataset,
batch_size,
shuffle=True,
drop_last=True)
# model
model = DeepLabV3()
model.cuda()
# define optimizer
optimizer = optim.Adam(model.parameters())
# define loss function
loss_fn = nn.CrossEntropyLoss()
# Metric
metric = SegmentationMetric(model.num_classes)
for epoch in range(100):
metric.reset()
for step, (input_tensor, label_tensor) in enumerate(data_loader):
import pickle
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import cv2
import time
# NOTE! NOTE! change this to not overwrite all log data when you train the model:
model_id = "1"
num_epochs = 100
batch_size = 8
learning_rate = 0.0001
network = DeepLabV3(model_id, project_dir=".").cuda()
train_dataset = DatasetTrain(cityscapes_data_path="./data/cityscapes",
cityscapes_meta_path="./data/cityscapes/meta")
val_dataset = DatasetVal(cityscapes_data_path="./data/cityscapes",
cityscapes_meta_path="./data/cityscapes/meta")
num_train_batches = int(len(train_dataset) / batch_size)
num_val_batches = int(len(val_dataset) / batch_size)
print("num_train_batches:", num_train_batches)
print("num_val_batches:", num_val_batches)
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=4)
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import cv2
import time
# NOTE! NOTE! change this to not overwrite all log data when you train the model:
model_id = "1"
num_epochs = 1000
batch_size = 3
learning_rate = 0.0001 #需要增加学习率的控制，不要固定一个学习率,应该迭代400时候降一次
network = DeepLabV3(
model_id,
project_dir="/workspace/mnt/group/other/luchao/deeplabv3/deeplabv3").cuda(
)
train_dataset = DatasetTrain(
cityscapes_data_path=
"/workspace/mnt/group/other/luchao/deeplabv3/deeplabv3/data/cityscapes",
cityscapes_meta_path=
"/workspace/mnt/group/other/luchao/deeplabv3/deeplabv3/data/cityscapes/meta"
)
val_dataset = DatasetVal(
cityscapes_data_path=
"/workspace/mnt/group/other/luchao/deeplabv3/deeplabv3/data/cityscapes",
cityscapes_meta_path=
"/workspace/mnt/group/other/luchao/deeplabv3/deeplabv3/data/cityscapes/meta"
)
for x in ["train", "validation"]
}
dataloaders = {
x: torch.utils.data.DataLoader(
image_datasets[x],
batch_size=batch_size,
#shuffle=x=="train",
sampler=data_sampler[x],
num_workers=2)
for x in ['train', 'validation']
}
use_gpu = torch.cuda.is_available()
model = DeepLabV3(model_id="1", project_dir=current)
model.load_state_dict(
torch.load(
os.path.join(current,
"pretrained_models/model_13_2_2_2_epoch_580.pth")))
model.aspp.conv_1x1_4 = nn.Conv2d(256, 5, kernel_size=1)
#model.load_state_dict()
"""
for idx,p in enumerate(model.parameters()):
if idx!=0:
p.requires_grad = False
"""
if use_gpu:
#torch.distributed.init_process_group(backend="nccl")
model = nn.DataParallel(model).to(device)
#model = model.cuda()
import torch.utils.data
import torch.nn as nn
from torch.autograd import Variable
import torch.optim as optim
import torch.nn.functional as F
import numpy as np
import pickle
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import cv2
batch_size = 2
network = DeepLabV3("eval_val", project_dir="/root/deeplabv3").cuda()
network.load_state_dict(torch.load("/root/deeplabv3/pretrained_models/model_13_2_2_2_epoch_580.pth"))
val_dataset = DatasetVal(cityscapes_data_path="/root/deeplabv3/data/cityscapes",
cityscapes_meta_path="/root/deeplabv3/data/cityscapes/meta")
num_val_batches = int(len(val_dataset)/batch_size)
print ("num_val_batches:", num_val_batches)
val_loader = torch.utils.data.DataLoader(dataset=val_dataset,
batch_size=batch_size, shuffle=False,
num_workers=1)
with open("/root/deeplabv3/data/cityscapes/meta/class_weights.pkl", "rb") as file: # (needed for python3)
class_weights = np.array(pickle.load(file))
class_weights = torch.from_numpy(class_weights)
from torch.autograd import Variable
import torch.optim as optim
import torch.nn.functional as F
import numpy as np
import pickle
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import cv2
import os
batch_size = 2 # very small
network = DeepLabV3("eval_seq", project_dir="/root/deeplabv3").cuda() # move the nn.Module into GPU CUDA.
network.load_state_dict(torch.load("/root/deeplabv3/pretrained_models/model_13_2_2_2_epoch_580.pth")) # load cityscapes trained weight
for sequence in ["00", "01", "02"]:
print(sequence)
val_dataset = DatasetSeq(cityscapes_data_path="/data/cityscapes_dataset",
cityscapes_meta_path="/data/cityscapes_dataset/meta",
sequence=sequence)
num_val_batches = int(len(val_dataset)/batch_size)
print ("num_val_batches:", num_val_batches)
val_loader = torch.utils.data.DataLoader(dataset=val_dataset,
batch_size=batch_size, shuffle=False,
num_workers=1)
import pickle
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import cv2
import time
# NOTE! NOTE! change this to not overwrite all log data when you train the model:
model_id = "1"
num_epochs = 1000
batch_size = 3
learning_rate = 0.0001
network = DeepLabV3(model_id, project_dir="/root/deeplabv3").cuda()
train_dataset = DatasetTrain(
cityscapes_data_path="/root/deeplabv3/data/cityscapes",
cityscapes_meta_path="/root/deeplabv3/data/cityscapes/meta")
val_dataset = DatasetVal(
cityscapes_data_path="/root/deeplabv3/data/cityscapes",
cityscapes_meta_path="/root/deeplabv3/data/cityscapes/meta")
num_train_batches = int(len(train_dataset) / batch_size)
num_val_batches = int(len(val_dataset) / batch_size)
print("num_train_batches:", num_train_batches)
print("num_val_batches:", num_val_batches)
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=batch_size,
print(model)
return model
def create_model(opt, segment_network):
model = get_model(opt.model)
instance = model(opt, segment_network)
print("Model [%s] was created" % type(instance).__name__)
return instance
if __name__ == '__main__':
with open('config.yml', 'r') as cfg_file:
opt = yaml.load(cfg_file)
opt = Namespace(**opt)
segment_network = DeepLabV3("eval_val", project_dir="drive_augumentation")
segment_network.load_state_dict(
torch.load("pretrained_models/model_13_2_2_2_epoch_580.pth",
map_location='cpu'))
segment_network.eval()
if len(opt.gpu_ids) > 1:
segment_network = segment_network.to('cuda:1')
else:
segment_network = segment_network.cuda('cuda:{}'.format(
opt.gpu_ids[0]))
model = create_model(opt, segment_network)
dataset = SegmentationUnpairedDataset(opt.root_path, opt.load_size,
opt.seg_size)
dataloader = torch.utils.data.DataLoader(dataset,
batch_size=opt.batch_size,
num_workers=8)
11: 24,
12: 25,
13: 26,
14: 27,
15: 28,
16: 31,
17: 32,
18: 33,
19: 0
}
trainId_to_id_map_func = np.vectorize(trainId_to_id.get)
batch_size = 2
model_id = 1
network = DeepLabV3("eval_val_for_metrics", project_dir=default_path).cuda()
#network.load_state_dict(torch.load(os.path.join(default_path,'pretrained_models/model_13_2_2_2_epoch_580.pth')))
#check last checkpoint
data_list = glob.glob(
os.path.join(
default_path, 'training_logs/model_' + str(model_id) +
'/checkpoints/model_' + str(model_id) + '_*.pth'))
#find latest checkpoint
latest_epoch = 0
for name in list(data_list):
if latest_epoch < int(getEpoch(name)):
latest_epoch = int(getEpoch(name))
if latest_epoch != 0:
network.load_state_dict(
torch.load(
# -*- coding: utf-8 -*- """ Created on Fri Mar 19 12:45:05 2021 @author: Admin """ import torch from load_data import load_data from learning_function import learning_function from torchsummary import summary from plot import plot from deeplabv3 import DeepLabV3 from config import config ##################################################################################################### ######################################## load data ################################################## ##################################################################################################### l_train = load_data(r"D:/MHS data segmentation labeling/data/l_train.npy") test = load_data(r"D:/MHS data segmentation labeling/data/test.npy") ##################################################################################################### #################################### student model ################################################## ##################################################################################################### model = DeepLabV3(config["number_classes"]) train_ious,test_ious = learning_function(model,l_train,test) plot(train_ious,test_ious)
