| | |
|---|
| | | # preprocess_config(kwargs) |
|---|
| | | # import pdb; pdb.set_trace() |
|---|
| | | # set random seed |
|---|
| | | registry_tables.print_register_tables() |
|---|
| | | registry_tables.print() |
|---|
| | | set_all_random_seed(kwargs.get("seed", 0)) |
|---|
| | | torch.backends.cudnn.enabled = kwargs.get("cudnn_enabled", torch.backends.cudnn.enabled) |
|---|
| | | torch.backends.cudnn.benchmark = kwargs.get("cudnn_benchmark", torch.backends.cudnn.benchmark) |
|---|
| | |
|---|
| | | frontend_class = registry_tables.frontend_classes.get(frontend.lower()) |
|---|
| | | frontend = frontend_class(**kwargs["frontend_conf"]) |
|---|
| | | kwargs["frontend"] = frontend |
|---|
| | | kwargs["input_size"] = frontend.output_size() |
|---|
| | | |
|---|
| | | # import pdb; |
|---|
| | | # pdb.set_trace() |
|---|
| | |
|---|
| | | # dataloader |
|---|
| | | batch_sampler = kwargs["dataset_conf"].get("batch_sampler", "DynamicBatchLocalShuffleSampler") |
|---|
| | | batch_sampler_class = registry_tables.batch_sampler_classes.get(batch_sampler.lower()) |
|---|
| | | batch_sampler = batch_sampler_class(dataset_tr, **kwargs.get("dataset_conf")) |
|---|
| | | if batch_sampler is not None: |
|---|
| | | batch_sampler = batch_sampler_class(dataset_tr, **kwargs.get("dataset_conf")) |
|---|
| | | dataloader_tr = torch.utils.data.DataLoader(dataset_tr, |
|---|
| | | collate_fn=dataset_tr.collator, |
|---|
| | | batch_sampler=batch_sampler, |
|---|
| | |
|---|
| | | pin_memory=True) |
|---|
| | | |
|---|
| | | |
|---|
| | | |
|---|
| | | trainer = Trainer( |
|---|
| | | model=model, |
|---|
| | | optim=optim, |
|---|
