NVIDIA DALI从入门到放弃之六：Geometric Transforms-CFANZ编程社区

NVIDIA DALI从入门到放弃之一：概述

NVIDIA DALI从入门到放弃之二：入门示例

NVIDIA DALI从入门到放弃之三：Data Loading

NVIDIA DALI从入门到放弃之四：Multiple GPU

NVIDIA DALI从入门到放弃之五：Image Processing

NVIDIA DALI从入门到放弃之六：Geometric Transforms

NVIDIA DALI从入门到放弃之七：Sequence Processing

NVIDIA DALI从入门到放弃之八：PyTorch Plugin API

1 Catalogue

rotation - rotate by given angle (in degrees) around given point and axis (for 3D only)
translation - translate by given offset
scale - scale by given factor
shear - shear by given factors or angles; there are 2 shear factors for 2D and 6 factors for 3D
crop - translates and scales so that input corners (from_start, from_end) map to output corners (to_start, to_end).

2 多个转换组合（Combinining Multiple Transforms）

2-1 passing transform matrix

pipe = dali.pipeline.Pipeline(6, 3, 0, seed = 1234)
with pipe:
    jpegs, _ = fn.file_reader(file_root=root_dir, files=image_files)
    images = fn.image_decoder(jpegs, device="mixed")

    size = encoded_images_sizes(jpegs)
    center = size / 2

    keypoints = fn.numpy_reader(file_root=root_dir, files=keypoint_files)
    mt = fn.transforms.rotation(angle = fn.random.uniform(range=(-45, 45)), center = center)
    mt = fn.transforms.translation(mt, offset = (400,0))
    images = fn.warp_affine(images, matrix = mt, fill_value=0, inverse_map=False)
    keypoints = fn.coord_transform(keypoints, MT = mt)
    pipe.set_outputs(images, keypoints)

pipe.build()
images, keypoints = pipe.run()

2-2 transforms.combine

pipe = dali.pipeline.Pipeline(6, 3, 0, seed = 1234)
with pipe:
    jpegs, _ = fn.file_reader(file_root=root_dir, files=image_files)
    images = fn.image_decoder(jpegs, device="mixed")

    size = encoded_images_sizes(jpegs)
    center = size / 2

    keypoints = fn.numpy_reader(file_root=root_dir, files=keypoint_files)
    tr1 = fn.transforms.translation(offset=-center)
    tr2 = fn.transforms.translation(offset=center)
    rot = fn.transforms.rotation(angle = fn.random.uniform(range=(-45, 45)))
    mt = fn.transforms.combine(tr1, rot, np.float32([[1, 1, 0],
                                               [0, 1, 0]]), tr2)
    images = fn.warp_affine(images, matrix = mt, fill_value=0, inverse_map=False)
    keypoints = fn.coord_transform(keypoints, MT = mt)
    pipe.set_outputs(images, keypoints)

pipe.build()
images, keypoints = pipe.run()

3 示例

pipe = dali.pipeline.Pipeline(6, 3, 0, seed = 1234)
with pipe:
    jpegs, _ = fn.file_reader(file_root=root_dir, files=image_files)
    images = fn.image_decoder(jpegs, device="mixed")
    keypoints = fn.numpy_reader(file_root=root_dir, files=keypoint_files)

    size = encoded_images_sizes(jpegs)
    center = size / 2

    outputs = []

    transforms = [
        fn.transforms.translation(offset = fn.random.uniform(range=(-100, 100), shape = 2)),
        fn.transforms.rotation(angle = fn.random.uniform(range=(-45, 45)), center = center),
        fn.transforms.scale(scale = fn.random.uniform(range=(0.5, 2), shape=[2]), center = center),
        fn.transforms.shear(shear = fn.random.uniform(range=(-1, 1), shape=[2]), center = center),
        fn.transforms.crop(from_start = size * 0.1, from_end = size * 0.8, to_start = [0,0], to_end=size*1.0)
    ]

    for mt in transforms:
        out_img = fn.warp_affine(images, matrix = mt, fill_value=0, inverse_map=False)
        out_kp = fn.coord_transform(keypoints, MT = mt)
        outputs += [out_img, out_kp]
    pipe.set_outputs(*outputs)

pipe.build()
pipe_out = pipe.run()