Crate tch_utils

Tch-utils - A collection of utilities for the tch-rs crate

Note : This crate is mostly intended for my own to collect some utilities I use in my projects. It is not intended to be used by anyone else. However, if you find it useful, feel free to use it.

Features

• 2D operations : operations on tensor of shape [N, C, H, W] on each pixel
• Tensor initialization : new way to initialize tensors
• Noises : Generate noise tensors

Conventions

Shapes

• N : The number of samples

• C : The number of channels

• H : The height of the image

• W : The width of the image

• [N, C, H, W] : A tensor of shape [N, C, H, W] is a batch of N images of shape [C, H, W] A tensor of multiple sample with one chanel will never be represented as [N, H, W] but as [N, 1, H, W]

• [C, H, W] : A tensor of shape [C, H, W] is an image of shape [C, H, W]

• [H, W] : A tensor of shape [H, W] is an image of shape [1, H, W]

tensor of 2d vectors Tensors of 2d vectors ([N, 2, H, W] & [2, H, W]) are in the form [y, x]

Axis

the y axis will always be top to bottom the x axis will always be left to right

0 ----> 1 (x)
|
|
v
1 
(y)

Colors

• Unless specified, the color channels are in the order [R, G, B]
• The color channels are in the dim 1 of [N, C, H, W] tensor
• Unless specified, the color channels are in the range [0, 1]
• Unless specified, the color channels use and expect Kind::Float

Modules

• color
• gabor
  Implementation of Gabor filter.
• glcm
• glrlm
  Gray Level Run Length Matrix (GLRLM).
• image
  Adds a trait to convert between tch::Tensor and image::DynamicImage.
• ndarray
• noise
  Noise
• ops_2d
  2D Operations
• shapes
  This module contains functions to generate shapes.
• tensor_ext
  Contains diverse extensions to the Tensor struct.
• tensor_init
  Tensor initialization functions
• utils
  Utilities for the utility crate :D