DL-Art-School/codes/data/__init__.py

88 lines
3.8 KiB
Python
Raw Normal View History

2019-08-23 13:42:47 +00:00
"""create dataset and dataloader"""
import logging
import torch
import torch.utils.data
from munch import munchify
2019-08-23 13:42:47 +00:00
2021-06-11 21:31:10 +00:00
from utils.util import opt_get
2019-08-23 13:42:47 +00:00
2021-08-17 04:52:15 +00:00
def create_dataloader(dataset, dataset_opt, opt=None, sampler=None, collate_fn=None, shuffle=True):
2019-08-23 13:42:47 +00:00
phase = dataset_opt['phase']
if phase == 'train':
2021-06-11 21:31:10 +00:00
if opt_get(opt, ['dist'], False):
2019-08-23 13:42:47 +00:00
world_size = torch.distributed.get_world_size()
num_workers = dataset_opt['n_workers']
assert dataset_opt['batch_size'] % world_size == 0
batch_size = dataset_opt['batch_size'] // world_size
else:
2021-06-11 21:31:10 +00:00
num_workers = dataset_opt['n_workers']
2019-08-23 13:42:47 +00:00
batch_size = dataset_opt['batch_size']
return torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=shuffle,
num_workers=num_workers, sampler=sampler, drop_last=True,
pin_memory=True, collate_fn=collate_fn)
2019-08-23 13:42:47 +00:00
else:
batch_size = dataset_opt['batch_size'] or 1
return torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=False, num_workers=0,
pin_memory=True, collate_fn=collate_fn)
2019-08-23 13:42:47 +00:00
def create_dataset(dataset_opt, return_collate=False):
2019-08-23 13:42:47 +00:00
mode = dataset_opt['mode']
collate = None
2019-08-23 13:42:47 +00:00
# datasets for image restoration
if mode == 'fullimage':
2020-08-25 17:56:59 +00:00
from data.full_image_dataset import FullImageDataset as D
2020-09-26 04:19:38 +00:00
elif mode == 'single_image_extensible':
from data.single_image_dataset import SingleImageDataset as D
2020-09-28 20:26:15 +00:00
elif mode == 'multi_frame_extensible':
from data.multi_frame_dataset import MultiFrameDataset as D
elif mode == 'combined':
from data.combined_dataset import CombinedDataset as D
2020-10-18 04:54:12 +00:00
elif mode == 'multiscale':
from data.multiscale_dataset import MultiScaleDataset as D
2020-10-24 02:58:07 +00:00
elif mode == 'paired_frame':
from data.paired_frame_dataset import PairedFrameDataset as D
2020-11-12 22:42:05 +00:00
elif mode == 'stylegan2':
from data.stylegan2_dataset import Stylegan2Dataset as D
elif mode == 'imagefolder':
from data.image_folder_dataset import ImageFolderDataset as D
2020-12-03 22:32:21 +00:00
elif mode == 'torch_dataset':
from data.torch_dataset import TorchDataset as D
elif mode == 'byol_dataset':
from data.byol_attachment import ByolDatasetWrapper as D
2020-12-10 22:07:35 +00:00
elif mode == 'byol_structured_dataset':
from data.byol_attachment import StructuredCropDatasetWrapper as D
elif mode == 'random_aug_wrapper':
from data.byol_attachment import DatasetRandomAugWrapper as D
2020-12-09 21:55:05 +00:00
elif mode == 'random_dataset':
from data.random_dataset import RandomDataset as D
2021-05-25 03:35:00 +00:00
elif mode == 'zipfile':
from data.zip_file_dataset import ZipFileDataset as D
elif mode == 'nv_tacotron':
from data.audio.nv_tacotron_dataset import TextMelLoader as D
from data.audio.nv_tacotron_dataset import TextMelCollate as C
from models.tacotron2.hparams import create_hparams
default_params = create_hparams()
2021-07-09 04:13:44 +00:00
default_params.update(dataset_opt)
dataset_opt = munchify(default_params)
2021-08-12 21:44:55 +00:00
if opt_get(dataset_opt, ['needs_collate'], True):
collate = C(dataset_opt.n_frames_per_step)
elif mode == 'gpt_tts':
from data.audio.gpt_tts_dataset import GptTtsDataset as D
from data.audio.gpt_tts_dataset import GptTtsCollater as C
collate = C(dataset_opt)
2021-09-14 23:43:16 +00:00
elif mode == 'unsupervised_audio':
from data.audio.unsupervised_audio_dataset import UnsupervisedAudioDataset as D
elif mode == 'unsupervised_audio_with_noise':
from data.audio.audio_with_noise_dataset import AudioWithNoiseDataset as D
2019-08-23 13:42:47 +00:00
else:
raise NotImplementedError('Dataset [{:s}] is not recognized.'.format(mode))
dataset = D(dataset_opt)
if return_collate:
return dataset, collate
else:
return dataset