VAL_TRANSFORMS = [
RandomResizedCrop(350, scale=(0.7, 1.0), interpolation=3),
RandomHorizontalFlip(p=0.5),
ToTensor(),
Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
]
BATCH_SIZE = 32
NUM_WORKERS = 15
dataset = FilesFromCsvDataset("output/filtered_train_dataset.csv")
targets = [y - 1 for _, y in dataset]
train_sampler = SmartWeightedRandomSampler(targets)
TRAIN_LOADER = get_data_loader(dataset,
data_transform=TRAIN_TRANSFORMS,
batch_size=BATCH_SIZE,
sampler=train_sampler,
num_workers=NUM_WORKERS,
cuda=True)
val_dataset = FilesFromCsvDataset("output/filtered_val_dataset.csv")
VAL_LOADER = get_data_loader(val_dataset,
data_transform=VAL_TRANSFORMS,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
cuda=True)
model_checkpoint = (
Path(OUTPUT_PATH) / "train_inceptionv4_350_smart_sampler_resized_crop" /
"20180502_0902" /
"model_FurnitureInceptionV4_350_1_val_loss=0.7726298.pth").as_posix()
MODEL = torch.load(model_checkpoint)
VAL_TRANSFORMS = [
RandomResizedCrop(350, scale=(0.8, 1.0)),
RandomHorizontalFlip(p=0.5),
ToTensor(),
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
]
BATCH_SIZE = 180
NUM_WORKERS = 15
dataset = FilesFromCsvDataset("output/filtered_train_dataset.csv")
TRAIN_LOADER = get_data_loader(dataset,
data_transform=TRAIN_TRANSFORMS,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
pin_memory=True)
val_dataset = FilesFromCsvDataset("output/filtered_val_dataset.csv")
VAL_LOADER = get_data_loader(val_dataset,
data_transform=VAL_TRANSFORMS,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
pin_memory=True)
MODEL = FurnitureSqueezeNet350(pretrained=True)
N_EPOCHS = 100
OPTIM = Adam(
he_map = dict([(v, i) for i, v in enumerate(hard_examples)])
def hard_examples_only(target):
return he_map[target] if target in he_map else len(hard_examples)
dataset = FilesFromCsvDataset("output/unique_filtered_train_dataset.csv")
train_sampler = get_weighted_train_sampler(dataset,
hard_examples_classes_weight,
n_samples=len(dataset))
TRAIN_LOADER = get_data_loader(dataset,
data_transform=TRAIN_TRANSFORMS,
target_transform=hard_examples_only,
sampler=train_sampler,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
pin_memory=True)
val_dataset = FilesFromCsvDataset("output/unique_filtered_val_dataset.csv")
val_sampler = get_weighted_train_sampler(val_dataset,
hard_examples_classes_weight,
n_samples=len(val_dataset))
VAL_LOADER = get_data_loader(val_dataset,
data_transform=VAL_TRANSFORMS,
target_transform=hard_examples_only,
sampler=val_sampler,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
pin_memory=True)
n_classes = 128
dataset = FilesFromCsvDataset("output/filtered_train_dataset.csv")
dataset = TransformedDataset(dataset,
transforms=read_image,
target_transforms=lambda l: l - 1)
dataset = TransformedDataset(dataset, transforms=Compose(TRAIN_TRANSFORMS))
dataset = RandomMultiImageAugDataset(dataset,
n_classes,
aug_fn=partial(basic_random_half_blend,
alpha=0.3))
dataset = TransformedDataset(dataset, transforms=Compose(common_transform))
TRAIN_LOADER = get_data_loader(dataset,
data_transform=None,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
cuda=True)
val_dataset = FilesFromCsvDataset("output/filtered_val_dataset.csv")
VAL_LOADER = get_data_loader(val_dataset,
data_transform=VAL_TRANSFORMS + common_transform,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
cuda=True)
MODEL = FurnitureSqueezeNet350(pretrained=True)
N_EPOCHS = 100
OPTIM = Adam(params=[
BATCH_SIZE = 32
NUM_WORKERS = 8
dataset = TrainvalFilesDataset(DATASET_PATH / "train_400x400")
dataset = TransformedDataset(dataset,
transforms=lambda x: x,
target_transforms=lambda y: y - 1)
reduced_train_indices = get_reduced_train_indices(dataset,
max_n_samples_per_class=1250,
seed=SEED)
del dataset
TRAIN_LOADER = get_data_loader(dataset_path=DATASET_PATH / "train_400x400",
data_transform=TRAIN_TRANSFORMS,
sample_indices=reduced_train_indices,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
cuda=True)
VAL_LOADER = get_data_loader(dataset_path=DATASET_PATH / "val_400x400",
data_transform=VAL_TRANSFORMS,
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
cuda=True)
MODEL = FurnitureSqueezeNet350(pretrained=True)
N_EPOCHS = 100
OPTIM = SGD(params=[
{
# Check training dataflow
from pathlib import Path
from torchvision.transforms import ToTensor
from common.data_loaders import get_data_loader
SEED = 12345
DEBUG = True
OUTPUT_PATH = "output/train_dataflow"
DATASET_PATH = Path("/home/fast_storage/imaterialist-challenge-furniture-2018/")
N_CLASSES = 128
N_EPOCHS = 1
BATCH_SIZE = 1
NUM_WORKERS = 8
DATA_LOADER = get_data_loader(
dataset_path=DATASET_PATH / "train",
data_transform=[ToTensor(), ],
batch_size=BATCH_SIZE,
num_workers=NUM_WORKERS,
cuda=False)
batch_size = 64
num_workers = 15
train_dataset = FilesFromCsvDataset("output/unique_filtered_train_dataset.csv")
val_dataset = FilesFromCsvDataset("output/unique_filtered_val_dataset.csv")
trainval_dataset = ConcatDataset([train_dataset, val_dataset])
# #### Stratified split :
fold_index = 2
n_splits = 4
train_index, val_index = get_trainval_indices(trainval_dataset,
fold_index=fold_index, n_splits=n_splits,
xy_transforms=None,
batch_size=batch_size, n_workers=8,
seed=SEED)
# ####
TEST_LOADER = get_data_loader(trainval_dataset,
data_transform=TEST_TRANSFORMS,
sample_indices=val_index,
batch_size=batch_size,
num_workers=num_workers,
pin_memory="cuda" in DEVICE)
MODEL = (Path("output") / "cv" / "nasnetlarge_350_resized_crop" / "fold_2" / "20180525_1653" /
"model_FurnitureNASNetALarge350_9_val_loss=0.4992649.pth").as_posix()
N_TTA = 7
