| 179 | |
| 180 | class DataModuleFromConfig(pl.LightningDataModule):# batchloader outputshape should be (b,h,w,c) and it will be permuted to (b,c,h,w) in autoencoder.get_input() |
| 181 | def __init__(self, batch_size, train=None, validation=None, test=None, predict=None, |
| 182 | wrap=False, num_workers=None, shuffle_test_loader=False, use_worker_init_fn=False, |
| 183 | shuffle_val_dataloader=False): |
| 184 | super().__init__() |
| 185 | self.batch_size = batch_size |
| 186 | self.dataset_configs = dict() |
| 187 | self.num_workers = num_workers if num_workers is not None else batch_size * 2 |
| 188 | self.use_worker_init_fn = use_worker_init_fn |
| 189 | if train is not None: |
| 190 | self.dataset_configs["train"] = train |
| 191 | self.train_dataloader = self._train_dataloader |
| 192 | if validation is not None: |
| 193 | self.dataset_configs["validation"] = validation |
| 194 | self.val_dataloader = partial(self._val_dataloader, shuffle=shuffle_val_dataloader) |
| 195 | if test is not None: |
| 196 | self.dataset_configs["test"] = test |
| 197 | self.test_dataloader = partial(self._test_dataloader, shuffle=shuffle_test_loader) |
| 198 | if predict is not None: |
| 199 | self.dataset_configs["predict"] = predict |
| 200 | self.predict_dataloader = self._predict_dataloader |
| 201 | self.wrap = wrap |
| 202 | |
| 203 | def prepare_data(self): |
| 204 | for data_cfg in self.dataset_configs.values(): |