iaa.Flipud(0.5),
iaa.Rot90((0, 4)),
# Blur and Noise
iaa.Sometimes(0.2, iaa.GaussianBlur(sigma=(0.25, 1.5), name="gaus-blur")),
iaa.Sometimes(
0.2,
iaa.AdditiveLaplaceNoise(scale=(0, 0.1 * 255),
per_channel=True,
name="gaus-noise")),
])
input_only = ["gaus-blur", "gaus-noise"]
db_train_list = []
for dataset in config.train.datasetsTrain:
db = dataloader.SurfaceNormalsDataset(input_dir=dataset.images,
label_dir=dataset.labels,
transform=augs_train,
input_only=input_only)
train_size = int(config.train.percentageDataForTraining * len(db))
db = torch.utils.data.Subset(db, range(train_size))
db_train_list.append(db)
db_train = torch.utils.data.ConcatDataset(db_train_list)
# Validation Dataset
augs_test = iaa.Sequential([
iaa.Resize({
"height": config.eval.imgHeight,
"width": config.eval.imgWidth
},
interpolation='nearest'),
])
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Enable Multi-GPU training
if torch.cuda.device_count() > 1:
print("Let's use", torch.cuda.device_count(), "GPUs!")
# dim = 0 [30, xxx] -> [10, ...], [10, ...], [10, ...] on 3 GPUs
net = nn.DataParallel(net)
augs_train = iaa.Sequential([
iaa.Scale((imsize, imsize), 0),
])
db_train = dataloader.SurfaceNormalsDataset(
input_dir='data/datasets/train/milk-bottles-train/resized-files/preprocessed-rgb-imgs',
label_dir='data/datasets/train/milk-bottles-train/resized-files/preprocessed-camera-normals',
transform=augs_train,
input_only=None,
)
trainLoader = DataLoader(db_train, batch_size=p['trainBatchSize'], shuffle=True, num_workers=32, drop_last=True)
# %matplotlib inline
lr_finder = LRFinder(net, optimizer, criterion, device="cuda")
lr_finder.range_test(trainLoader, end_lr=1, num_iter=100)
lr_finder.plot()
plt.show()
print(
colored(
'The dir to store results "{}" does not exist. Creating dir'.
format(DIR_RESULTS_SYNTHETIC), 'red'))
os.makedirs(DIR_RESULTS_SYNTHETIC)
###################### DataLoader #############################
print(colored('Will Run inference on these Test sets:', 'green'))
# Make new pytorch datasets for each synthetic dataset
db_test_list_synthetic = []
for dataset in config.eval.datasetsSynthetic:
print('Creating Synthetic Images dataset from: "{}"'.format(
dataset.images))
if dataset.images:
dataset = dataloader.SurfaceNormalsDataset(input_dir=dataset.images,
label_dir=dataset.labels,
transform=None,
input_only=None)
db_test_list_synthetic.append(dataset)
# Make new pytorch datasets for each real dataset
db_test_list_real = []
for dataset in config.eval.datasetsReal:
print('Creating Real Images dataset from: "{}"'.format(dataset.images))
if dataset.images:
dataset = dataloader.SurfaceNormalsRealImagesDataset(
input_dir=dataset.images,
imgHeight=config_checkpoint.train.imgHeight,
imgWidth=config_checkpoint.train.imgWidth)
db_test_list_real.append(dataset)
# Create pytorch dataloaders from datasets
