def main(args):
# Network Builders
builder = ModelBuilder()
net_encoder = builder.build_encoder(weights=args.weights_encoder)
net_decoder_1 = builder.build_decoder(weights=args.weights_decoder_1)
net_decoder_2 = builder.build_decoder(arch='c1',weights=args.weights_decoder_2)
if args.weighted_class:
crit = nn.NLLLoss(ignore_index=-1, weight=args.class_weight)
else:
crit = nn.NLLLoss(ignore_index=-1)
# Dataset and Loader
dataset_train = GTA(root=args.root_gta, cropSize=args.imgSize, is_train=1)
dataset_val = CityScapes('val', root=args.root_cityscapes, cropSize=args.imgSize,
max_sample=args.num_val, is_train=0)
dataset_val_2 = BDD('val', root=args.root_bdd, cropSize=args.imgSize,
max_sample=args.num_val, is_train=0)
loader_val = torch.utils.data.DataLoader(
dataset_val,
batch_size=args.batch_size_eval,
shuffle=False,
num_workers=int(args.workers),
drop_last=True)
loader_val_2 = torch.utils.data.DataLoader(
dataset_val_2,
batch_size=args.batch_size_eval,
shuffle=False,
num_workers=int(args.workers),
drop_last=True)
args.epoch_iters = int(len(dataset_train) / args.batch_size)
print('1 Epoch = {} iters'.format(args.epoch_iters))
# load nets into gpu
if args.num_gpus > 1:
net_encoder = nn.DataParallel(net_encoder,
device_ids=range(args.num_gpus))
net_decoder_1 = nn.DataParallel(net_decoder_1,
device_ids=range(args.num_gpus))
net_decoder_2 = nn.DataParallel(net_decoder_2,
device_ids=range(args.num_gpus))
nets = (net_encoder, net_decoder_1, net_decoder_2, crit)
for net in nets:
net.cuda()
history = {split: {'epoch': [], 'err': [], 'acc': [], 'mIoU': []}
for split in ('train', 'val', 'val_2')}
# eval
evaluate(nets, loader_val, loader_val_2, history, 0, args)
print('Evaluation Done!')
def main():
colors = loadmat('colormap.mat')['colors']
dataset = GTA(root='/home/selfdriving/datasets/GTA_full', is_train=0)
h_s, w_s = 720, 1312
#dataset = CityScapes('val', root='/home/selfdriving/datasets/cityscapes_full', is_train=0)
#h_s, w_s = 720, 1440
#dataset = BDD('val',root='/home/selfdriving/datasets/bdd100k', is_train=0)
#h_s, w_s = 720, 1280
loader = torch.utils.data.DataLoader(dataset,
batch_size=16,
shuffle=True,
num_workers=8,
drop_last=False)
count = 0
statistics = np.zeros([19], dtype=np.int)
mean_image = torch.zeros([19, h_s, w_s]).to(torch.long)
for batch_data in tqdm(loader):
(imgs, segs, infos) = batch_data
count += imgs.size(0)
for i in range(19):
statistics[i] += torch.sum(segs == i)
mean_image[i] += torch.sum(segs == i, dim=0)
pred = mean_image.data.cpu().numpy() / count
pred_ = np.argmax(pred, axis=0)
pred_color = colorEncode(pred_, colors).astype(np.uint8)
#print(entropy(np.ones(19)))
entropies = entropy(pred)
plt.imshow(entropies, cmap='hot')
plt.colorbar()
plt.savefig('./entropies.png')
plt.clf()
max_vals = np.max(pred, axis=0)
plt.imshow(max_vals, cmap='hot')
plt.colorbar()
plt.savefig('./max_vals.png')
imsave('./mean.png', pred_color)
def main(args):
# Network Builders
builder = ModelBuilder()
net_encoder = builder.build_encoder(weights=args.weights_encoder)
net_decoder_1 = builder.build_decoder(weights=args.weights_decoder_1)
net_decoder_2 = builder.build_decoder(arch='c1',
weights=args.weights_decoder_2)
if args.weighted_class:
crit = nn.NLLLoss(ignore_index=-1, weight=args.class_weight)
else:
crit = nn.NLLLoss(ignore_index=-1)
# Dataset and Loader
dataset_train = GTA(root=args.root_gta,
cropSize=args.imgSize,
is_train=1,
random_mask=args.mask)
dataset_val = CityScapes('val',
root=args.root_cityscapes,
cropSize=args.imgSize,
max_sample=args.num_val,
is_train=0)
dataset_val_2 = BDD('val',
root=args.root_bdd,
cropSize=args.imgSize,
max_sample=args.num_val,
is_train=0)
loader_train = torch.utils.data.DataLoader(dataset_train,
batch_size=args.batch_size,
shuffle=True,
num_workers=int(args.workers),
drop_last=True)
loader_val = torch.utils.data.DataLoader(dataset_val,
batch_size=args.batch_size_eval,
shuffle=False,
num_workers=int(args.workers),
drop_last=True)
loader_val_2 = torch.utils.data.DataLoader(dataset_val_2,
batch_size=args.batch_size_eval,
shuffle=False,
num_workers=int(args.workers),
drop_last=True)
args.epoch_iters = int(len(dataset_train) / args.batch_size)
print('1 Epoch = {} iters'.format(args.epoch_iters))
# load nets into gpu
if args.num_gpus > 1:
net_encoder = nn.DataParallel(net_encoder,
device_ids=range(args.num_gpus))
net_decoder_1 = nn.DataParallel(net_decoder_1,
device_ids=range(args.num_gpus))
net_decoder_2 = nn.DataParallel(net_decoder_2,
device_ids=range(args.num_gpus))
nets = (net_encoder, net_decoder_1, net_decoder_2, crit)
for net in nets:
net.cuda()
# Set up optimizers
optimizers = create_optimizers(nets, args)
# Main loop
history = {
split: {
'epoch': [],
'err': [],
'acc': [],
'mIoU': []
}
for split in ('train', 'val', 'val_2')
}
# optional initial eval
# evaluate(nets, loader_val, loader_val_2, history, 0, args)
for epoch in range(1, args.num_epoch + 1):
train(nets, loader_train, optimizers, history, epoch, args)
# Evaluation
if epoch % args.eval_epoch == 0:
evaluate(nets, loader_val, loader_val_2, history, epoch, args)
# checkpointing
checkpoint(nets, history, args)
# adjust learning rate
adjust_learning_rate(optimizers, epoch, args)
print('Training Done!')
def main(args):
# Network Builders
builder = ModelBuilder()
net_encoder = builder.build_encoder(weights=args.weights_encoder)
net_decoder_1 = builder.build_decoder(weights=args.weights_decoder_1,
use_softmax=True)
net_decoder_2 = builder.build_decoder(weights=args.weights_decoder_2,
use_softmax=True)
if args.weighted_class:
crit1 = nn.CrossEntropyLoss(ignore_index=-1, weight=args.class_weight)
else:
crit1 = nn.CrossEntropyLoss(ignore_index=-1)
# Dataset and Loader
dataset_train_sup = GTA(root=args.root_gta,
cropSize=args.imgSize,
max_sample=-1,
is_train=1,
random_mask=args.mask)
split_lengths = [len(dataset_train_sup) // args.gamma1] * args.gamma1
split_lengths[0] += len(dataset_train_sup) % args.gamma1
split_dataset_train_sup = torch.utils.data.random_split(
dataset_train_sup, split_lengths)
dataset_train_unsup = CityScapes('train',
root=args.root_cityscapes,
cropSize=args.imgSize,
max_sample=-1,
is_train=1)
split_lengths = [len(dataset_train_unsup) // args.gamma2] * args.gamma2
split_lengths[0] += len(dataset_train_unsup) % args.gamma2
split_dataset_train_unsup = torch.utils.data.random_split(
dataset_train_unsup, split_lengths)
dataset_val = CityScapes('val',
root=args.root_cityscapes,
cropSize=args.imgSize,
max_sample=args.num_val,
is_train=0)
loader_val = torch.utils.data.DataLoader(dataset_val,
batch_size=args.batch_size_eval,
shuffle=False,
num_workers=int(args.workers),
drop_last=True)
args.epoch_iters = int(
(len(dataset_train_sup) // args.gamma1 +
len(dataset_train_unsup) // args.gamma2) / args.batch_size)
print('1 Epoch = {} iters'.format(args.epoch_iters))
# load nets into gpu
if args.num_gpus > 1:
net_encoder = nn.DataParallel(net_encoder,
device_ids=range(args.num_gpus))
net_decoder_1 = nn.DataParallel(net_decoder_1,
device_ids=range(args.num_gpus))
net_decoder_2 = nn.DataParallel(net_decoder_2,
device_ids=range(args.num_gpus))
nets = (net_encoder, net_decoder_1, net_decoder_2, crit1)
for net in nets:
net.cuda()
# Set up optimizers
optimizers = create_optimizers(nets, args)
# Main loop
history = {
split: {
'epoch': [],
'err': [],
'acc': [],
'mIoU': []
}
for split in ('train', 'val')
}
# optional initial eval
# evaluate(nets, loader_val, history, 0, args)
for epoch in range(1, args.num_epoch + 1):
train(nets, split_dataset_train_sup, split_dataset_train_unsup,
optimizers, history, epoch, args)
# Evaluation and visualization
if epoch % args.eval_epoch == 0:
evaluate(nets, loader_val, history, epoch, args)
# checkpointing
checkpoint(nets, history, args)
# adjust learning rate
adjust_learning_rate(optimizers, epoch, args)
print('Training Done!')