def create_dali_pipeline()

in src/sm_augmentation_train-script.py [0:0]

• 27 lines of code
• 5 McCabe index (conditional complexity)

def create_dali_pipeline(data_dir, crop, size, shard_id, num_shards, dali_cpu, is_training, aug_load_factor):
    images, labels = fn.readers.file(file_root=data_dir,
                                     shard_id=shard_id,
                                     num_shards=num_shards,
                                     random_shuffle=is_training,
                                     pad_last_batch=True,
                                     name="Reader")
    """
    For jpeg images, “mixed” backend uses the nvJPEG library. If hardware is available, operator will use dedicated 
    hardware decoder. For jpeg images, “cpu” backend uses libjpeg-turbo. Other image formats are decoded with OpenCV 
    or other specific libraries, such as libtiff. 
    """
    dali_device = 'cpu' if dali_cpu else 'gpu'
    decoder_device = 'cpu' if dali_cpu else 'mixed'

    images = fn.decoders.image(images,
                               device=decoder_device,
                               output_type=types.RGB,
                               memory_stats=True)
    if is_training:
        # Repeating Augmentation to influence bottleneck
        for x in range(aug_load_factor):
            images = fn.flip(images, device=dali_device, horizontal=1, vertical=1)
            images = fn.rotate(images, angle=5, device=dali_device)

    images = fn.random_resized_crop(images, size=size, device=dali_device)
    images = fn.crop_mirror_normalize(images,
                                      dtype=types.FLOAT,
                                      output_layout="CHW",
                                      crop=(crop, crop),
                                      mean=[0.485 * 255, 0.456 * 255, 0.406 * 255],
                                      std=[0.229 * 255, 0.224 * 255, 0.225 * 255])
    images = images.gpu()
    labels = labels.gpu()

    return images, labels