boostcampaitech3 / level1-image-classification-level1-nlp-10

0 stars 4 forks source link

Feature/augmentation #36

Closed hyoeun98 closed 2 years ago

hyoeun98 commented 2 years ago

변경점

dataset.py


best model config.json

{"seed": 42, 
"epochs": 20, 
"dataset": "MaskSplitByProfileDataset", 
"augmentation": "CustomAugmentation", 
"resize": [256, 192], 
"batch_size": 64,
 "valid_batch_size": 64,
 "model": "densenet",
 "optimizer": "AdamW",
 "lr": 9e-05,
 "val_ratio": 0.2,
 "criterion": "focal",
 "lr_decay_step": 1,
 "log_interval": 20, 
"name": "exp", 
"weight_decay": 0.0005,
 "pretrained": "True", 
"early_stop": 1, 
"data_dir": "/opt/ml/input/data/train/images",
 "model_dir": "./model",
 "LR_scheduler": "StepLR",
 "precision": "True",
 "KfoldCV": "False", 
"early stop epoch": 2,
 "transform": "Compose(\n    Resize(size=[256, 192], interpolation=PIL.Image.BILINEAR)\n    RandomHorizontalFlip(p=0.5)\n    ToTensor()\n    Normalize(mean=(0.548, 0.504, 0.479), std=(0.237, 0.247, 0.246))\n)"}
hyoeun98 commented 2 years ago

변경점

def get_scheduler(optimizer):
    ...
    return scheduler

def get_dataset():
    ...
    return dataset

def get_transform(dataset):
    ...
    return transform

def get_loss_optim(model):
    ...
    return criterion, optimizer

def get_logger(save_dir)
    ...
    return logger