Open Dongyu-Han opened 1 week ago
def prepare_testing_data(config): def get_test_data_loader(config, test_name): # update the config dictionary with the specific testing dataset config = config.copy() # create a copy of config to avoid altering the original one config['test_dataset'] = test_name # specify the current test dataset if not config.get('dataset_type', None) == 'lrl': test_set = DeepfakeAbstractBaseDataset( config=config, mode='test', ) else: test_set = LRLDataset( config=config, mode='test', ) test_data_loader = \ torch.utils.data.DataLoader( dataset=test_set, batch_size=config['test_batchSize'], shuffle=False, num_workers=int(config['workers']), collate_fn=test_set.collate_fn, **drop_last = (test_name=='DeepFakeDetection'),** ) return test_data_loader test_data_loaders = {} for one_test_name in config['test_dataset']: test_data_loaders[one_test_name] = get_test_data_loader(config, one_test_name) return test_data_loaders
Why do you discard the remaining data of DeepFakeDetection? This brings about a mismatch in data size in the evaluation. for example:
Will removing this solve the problem?
Why do you discard the remaining data of DeepFakeDetection? This brings about a mismatch in data size in the evaluation. for example: