Augmentations

packnet_sfm.datasets.augmentations.colorjitter_sample(sample, parameters, prob=1.0)[source]

Jitters input images as data augmentation.

Parameters

  • sample (dict) – Input sample

  • parameters (tuple (brightness, contrast, saturation, hue)) – Color jittering parameters

  • prob (float) – Jittering probability

Returns

sample – Jittered sample

Return type

dict

packnet_sfm.datasets.augmentations.duplicate_sample(sample)[source]

Duplicates sample images and contexts to preserve their unaugmented versions.

Parameters

sample (dict) – Input sample

Returns

sample – Sample including [+”_original”] keys with copies of images and contexts.

Return type

dict

packnet_sfm.datasets.augmentations.resize_depth(depth, shape)[source]

Resizes depth map.

Parameters

  • depth (np.array [h,w]) – Depth map

  • shape (tuple (H,W)) – Output shape

Returns

depth – Resized depth map

Return type

np.array [H,W]

packnet_sfm.datasets.augmentations.resize_image(image, shape, interpolation=1)[source]

Resizes input image.

Parameters

  • image (Image.PIL) – Input image

  • shape (tuple [H,W]) – Output shape

  • interpolation (int) – Interpolation mode

Returns

image – Resized image

Return type

Image.PIL

packnet_sfm.datasets.augmentations.resize_sample(sample, shape, image_interpolation=1)[source]

Resizes a sample, including image, intrinsics and depth maps.

Parameters

  • sample (dict) – Dictionary with sample values

  • shape (tuple (H,W)) – Output shape

  • image_interpolation (int) – Interpolation mode

Returns

sample – Resized sample

Return type

dict

packnet_sfm.datasets.augmentations.resize_sample_image_and_intrinsics(sample, shape, image_interpolation=1)[source]

Resizes the image and intrinsics of a sample

Parameters

  • sample (dict) – Dictionary with sample values

  • shape (tuple (H,W)) – Output shape

  • image_interpolation (int) – Interpolation mode

Returns

sample – Resized sample

Return type

dict

packnet_sfm.datasets.augmentations.to_tensor(image, tensor_type='torch.FloatTensor')[source]

Casts an image to a torch.Tensor

packnet_sfm.datasets.augmentations.to_tensor_sample(sample, tensor_type='torch.FloatTensor')[source]

Casts the keys of sample to tensors.

Parameters

  • sample (dict) – Input sample

  • tensor_type (str) – Type of tensor we are casting to

Returns

sample – Sample with keys cast as tensors

Return type

dict