import torch
from torch import nn
import pretrainedmodels
from torchvision import transforms
from torch.utils.data import DataLoader
from src.utils.misc import AttrDict
from src.data.storms_dataset import get_storms_df, StormsDatasetRGBSequence, DATASET_MEAN, DATASET_STD
from src.data.data_feeders import DataFeed
MODEL_TYPE = 'L1A'
MODEL_VERSION = 'C'
TRAINING_VERSION = '2' # se_resnext101_32x4d_s03g10_v2 - B01n06
args = AttrDict({})
# Data
args.update({
'sequence_gap': 1, # hours
})
# Model
args.update({
'basemodel_name': 'se_resnext101_32x4d',
'pretrained': 'imagenet', # 'imagenet' or 'imagenet+background'
'freeze_basemodel':
False, # Indicate first layer to train, True for freeze all, False none
'reset_weights': False,
})
# Transformations
SIZE = 224
import torch
from torch import nn
import pretrainedmodels
from torchvision import transforms
from torch.utils.data import DataLoader
from src.utils.misc import AttrDict
from src.data.storms_dataset import get_storms_df, StormsDatasetSequence, DATASET_MEAN, DATASET_STD
from src.data.data_feeders import DataFeed
from src.pytorch.transformer import TransformerEncoder
MODEL_TYPE = 'L1A'
MODEL_VERSION = 'A'
TRAINING_VERSION = '4' # vgg11_transformer_s24g05B - CS061tbn24g05r3
args = AttrDict({})
# Data
args.update({
'sequence_gap': 0.5, # hours
'sequence_length': 8 * 3,
'train_split_ratio': 0.40,
})
# Train ALL
args.update({
'all_train_epochs': None,
'all_train_lr_step': None,
})
# Model
args.update({
'basemodel_name': 'vgg11_bn',
import torch
from torch import nn
from torch.utils.data import DataLoader
from torchvision import transforms
from src.data.geodata_utils import get_data_dict, get_dataset_df
from src.utils.misc import AttrDict
from src.data.roof_dataset import RoofDataset
from src.utils.image_transformations import trim
from src.data.data_feeders import DataFeed
MODEL_TYPE = 'L1A'
MODEL_VERSION = 'C'
TRAINING_VERSION = '06g'
args = AttrDict({})
# Data
args.update({
'use_only_verified':
True, # Whether to use only verified data to train the model
'add_location': False,
'train_split_ratio': 0.091, # Ratio split train/validation (if needed)
'random_split_seed': 999, # Random seed for splitting training set
'correct_train_samples': None,
'thershold_for_corrections': None, # None or float
'validate_with_corrected_labels':
False, # Use original labels or corrected to validate
})
# Model
args.update({
from torch import nn
import pretrainedmodels
from torchvision import transforms
from torch.utils.data import DataLoader
from src.utils.misc import AttrDict
from src.data.storms_dataset import get_storms_df, StormsDatasetSequence, DATASET_MEAN, DATASET_STD
from src.data.data_feeders import DataFeed
from src.pytorch.transformations import RandomRingMask, RandomCircleMask
MODEL_TYPE = 'L1A'
MODEL_VERSION = 'C'
TRAINING_VERSION = '3' # se_resnext101_32x4d_s07g10 - C01
args = AttrDict({})
# Data
args.update({
'sequence_gap': 1, # hours
'sequence_length': 7,
})
# Train ALL
args.update({
'all_train_epochs': None,
'all_train_lr_step': None,
})
# Model
args.update({
'basemodel_name': 'se_resnext101_32x4d',
'pretrained': 'imagenet', # 'imagenet' or 'imagenet+background'
import torch
from torch import nn
from torch.utils.data import DataLoader
from torchvision import transforms
from src.data.geodata_utils import get_data_dict, get_dataset_df
from src.utils.misc import AttrDict
from src.data.roof_dataset import RoofDataset
from src.utils.image_transformations import trim
from src.data.data_feeders import DataFeed
MODEL_TYPE = 'L1A'
MODEL_VERSION = 'A'
TRAINING_VERSION = '06'
args = AttrDict({})
# Data
args.update({
'use_only_verified':
True, # Whether to use only verified data to train the model
'train_split_ratio': 0.091, # Ratio split train/validation (if needed)
'random_split_seed': 999, # Random seed for splitting training set
})
# Model
args.update({
'basemodel_name': 'dpn92',
'pretrained':
'imagenet+5k', # 'imagenet' or 'imagenet+background' or 'imagenet+5k'
'freeze_basemodel':
False, # Indicate first layer to train, True for freeze all, False none