Transforms Utils

class master_thesis.utils.TransformsUtils

Bases: object

static resize(image, size, mode='bilinear', keep_ratio=True)

Resize an image using the the algorithm given in mode.

Parameters

  • image (torch.FloatTensor) – tensor of size (C,F,H,W) containing the image quantized from [0, 1].

  • size (tuple) – tuple containing the desired size in the form (H, W).

  • mode (str) – mode used to resize the image. Same format as in torch.nn.functional.interpolate().

Returns

resized image.

Return type

torch.FloatTensor

static resize_set(x, v, y, size)

Resizes the entire set of data (x, m, y).

Parameters

  • x (torch.FloatTensor) – tensor of size (B,C,F,H,W) containing masked

  • [0 (from) –

  • 1] –

  • v (torch.FloatTensor) – tensor of size (B,1,F,H,W) containing visibility

  • [0 –

  • 1] –

  • y (torch.FloatTensor) – tensor of size (B,C,F,H,W) containing images

  • [0 –

  • 1] –

  • size (int) – new size of the set of data.

static crop(image, size, crop_center=True, crop_position=None)

Crop a patch from the image.

Parameters

  • image (torch.FloatTensor) – tensor of size (C, F, H, W) containing

  • image. (the) –

  • size (tuple) – tuple containing the desired size in the form (H, W).

  • crop_position (tuple) – coordinates of the top-left pixel from where

  • set (to cut the patch. If not) –

  • randomly. (it is generated) –

Returns

patch of the image.

Return type

torch.FloatTensor

static dilatate(images, filter_size, iterations)

Dilatates an image with a filter of size filter_size.

Parameters

  • images (torch.FloatTensor) – tensor of size (1,F,H,W) containing

  • image. (the) –

  • filter_size (tuple) – size of the filter in the form (H,W).

  • iterations (integer) – number of times to apply the filter.

Returns

dilatated image.

Return type

torch.FloatTensor

static interpolate_data(x_target, m_target, x_ref, m_ref, h_new, w_new)