habana_frameworks.mediapipe.fn.Transpose

Class:

• habana_frameworks.mediapipe.fn.Transpose(**kwargs)

Define graph call:

• __call__(input)

Parameter:

• input - Input tensor to operator. Supported dimensions: minimum = 1, maximum = 5. Supported data types: INT8, UINT8, BFLOAT16, FLOAT32.

Description:

Produces a transposed tensor of the input tensor along multiple axes. See https://github.com/onnx/onnx/blob/rel-1.9.0/docs/Operators.md#Transpose.

Supported backend:

• HPU

Keyword Arguments

kwargs	Description
permutation	Each location in the permutation array defines to which dimension this location will be transposed to, permutation[rank(Input tensor)], Type: list[int] Default: [0, 1, 2, 3, 4] Optional: yes
tensorDim	Number of valid entries in the permutation array. Type: int Default: 5 Optional: no
dtype	Output data type. Type: habana_frameworks.mediapipe.media_types.dtype Default: UINT8 Optional: yes Supported data types: INT8 UINT8 BFLOAT16 FLOAT32

Note

1. Input/output tensors must be of the same datatype.

2. Input/output tensor must have same number of dimensions and total elements.

Example: Transpose Operator

The following code snippet shows usage of Transpose operator:

from habana_frameworks.mediapipe import fn
from habana_frameworks.mediapipe.mediapipe import MediaPipe
from habana_frameworks.mediapipe.media_types import imgtype as it
from habana_frameworks.mediapipe.media_types import dtype as dt
import matplotlib.pyplot as plt
import numpy as np
import os

g_display_timeout = os.getenv("DISPLAY_TIMEOUT") or 5


# Create MediaPipe derived class
class myMediaPipe(MediaPipe):
    def __init__(self, device, queue_depth, batch_size, num_threads, op_device, dir, img_h, img_w):
        super(
            myMediaPipe,
            self).__init__(
            device,
            queue_depth,
            batch_size,
            num_threads,
            self.__class__.__name__)

        self.input = fn.ReadImageDatasetFromDir(shuffle=False,
                                                dir=dir,
                                                format="jpg",
                                                device="cpu")

        # WHCN
        self.decode = fn.ImageDecoder(device="hpu",
                                      output_format=it.RGB_I,
                                      resize=[img_w, img_h])

        # WHCN -> CWHN
        self.transpose = fn.Transpose(permutation=[0, 2, 1, 3],
                                      tensorDim=4,
                                      dtype=dt.UINT8,
                                      device=op_device)

    def definegraph(self):
        imgs, lbls = self.input()
        imgs = self.decode(imgs)
        transp_imgs = self.transpose(imgs)
        return transp_imgs, imgs, lbls


def display_images(images, batch_size, cols):
    rows = (batch_size + 1) // cols
    plt.figure(figsize=(10, 10))
    for i in range(batch_size):
        ax = plt.subplot(rows, cols, i + 1)
        plt.imshow(images[i])
        plt.axis("off")
    plt.show(block=False)
    plt.pause(g_display_timeout)
    plt.close()

def run(device, op_device):
    batch_size = 6
    queue_depth = 2
    num_threads = 1
    img_width = 200
    img_height = 200
    base_dir = os.environ['DATASET_DIR']
    dir = base_dir + "/img_data/"
    columns = 3

    # Create MediaPipe object
    pipe = myMediaPipe(device, queue_depth, batch_size,
                      num_threads, op_device, dir,
                      img_height, img_width)

    # Build MediaPipe
    pipe.build()

    # Initialize MediaPipe iterator
    pipe.iter_init()

    # Run MediaPipe
    transp_imgs, imgs, lbls = pipe.run()

    def as_cpu(tensor):
        if (callable(getattr(tensor, "as_cpu", None))):
            tensor = tensor.as_cpu()
        return tensor

    # Copy data to host from device as numpy array
    transp_imgs = as_cpu(transp_imgs).as_nparray()
    imgs = as_cpu(imgs).as_nparray()
    lbls = as_cpu(lbls).as_nparray()

    del pipe

    # Display shape
    print('input shape', imgs.shape)
    print('output shape', transp_imgs.shape)
    display_images(transp_imgs, batch_size, columns)
    return imgs, transp_imgs


def compare_ref(inp, out):
    ref = np.transpose(inp, [0, 2, 1, 3])
    if np.array_equal(ref, out) == False:
        raise ValueError(f"Mismatch w.r.t ref")


if __name__ == "__main__":
    dev_opdev = {'mixed': ['hpu'],
                'legacy': ['hpu']}

    for dev in dev_opdev.keys():
        for op_dev in dev_opdev[dev]:
            inp, out = run(dev, op_dev)
            compare_ref(inp, out)

Transposed Output Images 1

1

Licensed under a CC BY SA 4.0 license. The images used here are taken from https://data.caltech.edu/records/mzrjq-6wc02.

Images shape before transpose operation: NCHW

image shape: (6, 3, 200, 200)

Images shape after transpose operation: NHWC

image shape: (6, 200, 200, 3)