Transform inputs for decoder. Args: inputs (list[Tensor]): List of multi-level img features. Returns: Tensor:

Add prefix for dict. Args: inputs (dict): The input dict with str keys. prefix (str): The prefix to add. Returns: d

Test with augmentations. Only rescale=True is supported.

x_list contains a list of tensors to nest together and run

x_list contains a list of tensors to nest together and run

Load and resize a depth map.

Load and resize an image.

Inverse cumulative distribution function of the Gaussian distribution.

Apply activation functions for dynamic keys.

Loads weights from the trained decoder checkpoint.

Log probability density function of the Gaussian distribution.

Compute depth loss.

Resize a PIL image so that its height and width are the closest multiples of 14. The resizing is done with bilinear interpolation.

Resize the input PIL image so that its shorter side equals target_size (default 518) while keeping the aspect ratio.

Simple test with single image.

Inference by sliding-window with overlap. If h_crop > h_img or w_crop > w_img, the small patch will be used to decode without padding

Inference with full image.