def build_dataloader()

in utils/main_utils.py [0:0]

• 73 lines of code
• 8 McCabe index (conditional complexity)

def build_dataloader(db_cfg, split_cfg, num_workers, distributed):
    import torch.utils.data as data
    import torch.utils.data.distributed
    from datasets import preprocessing
    import datasets

    # Video transforms
    num_frames = int(db_cfg['video_clip_duration'] * db_cfg['video_fps'])
    if db_cfg['transforms'] == 'crop+color':
        video_transform = preprocessing.VideoPrep_Crop_CJ(
            resize=db_cfg['frame_size'],
            crop=(db_cfg['crop_size'], db_cfg['crop_size']),
            augment=split_cfg['use_augmentation'],
            num_frames=num_frames,
            pad_missing=True,
        )

    elif db_cfg['transforms'] == 'msc+color':
        video_transform = preprocessing.VideoPrep_MSC_CJ(
            crop=(db_cfg['crop_size'], db_cfg['crop_size']),
            augment=split_cfg['use_augmentation'],
            num_frames=num_frames,
            pad_missing=True,
        )

    else:
        raise ValueError('Unknown transform')

    # Audio transforms
    audio_transforms = [
        preprocessing.AudioPrep(
            trim_pad=True,
            duration=db_cfg['audio_clip_duration'],
            augment=split_cfg['use_augmentation'],
            missing_as_zero=True),
        preprocessing.LogSpectrogram(
            db_cfg['audio_fps'],
            n_fft=db_cfg['n_fft'],
            hop_size=1. / db_cfg['spectrogram_fps'],
            normalize=True)
    ]
    audio_fps_out = db_cfg['spectrogram_fps']

    if db_cfg['name'] == 'audioset':
        dataset = datasets.AudioSet
    elif db_cfg['name'] == 'kinetics':
        dataset = datasets.Kinetics
    else:
        raise ValueError('Unknown dataset')

    clips_per_video = split_cfg['clips_per_video'] if 'clips_per_video' in split_cfg else 1
    db = dataset(
        subset=split_cfg['split'],
        return_video=True,
        video_clip_duration=db_cfg['video_clip_duration'],
        video_fps=db_cfg['video_fps'],
        video_transform=video_transform,
        return_audio=True,
        audio_clip_duration=db_cfg['audio_clip_duration'],
        audio_fps=db_cfg['audio_fps'],
        audio_fps_out=audio_fps_out,
        audio_transform=audio_transforms,
        max_offsync_augm=0.5 if split_cfg['use_augmentation'] else 0,
        return_labels=False,
        return_index=True,
        mode='clip',
        clips_per_video=clips_per_video,
    )

    if distributed:
        sampler = torch.utils.data.distributed.DistributedSampler(db)
    else:
        sampler = None

    loader = torch.utils.data.DataLoader(
        db,
        batch_size=db_cfg['batch_size'],
        shuffle=(sampler is None),
        drop_last=split_cfg['drop_last'],
        num_workers=num_workers,
        pin_memory=True,
        sampler=sampler)

    return loader