Create a list of training samples based on an image and a mask. Args: mat: 3D hyperspectral matrix to extract the spectrums from

Instantiate and obtain a model with adequate hyperparameters Args: name: string of the model name kwargs: hyperparameters

embed_dim: output dimension for each position pos: a list of positions to be encoded: size (M,) out: (M, D)

if fix_train_patches: fix patches that will be used for training during initialization if False, patches will be drawn randomly from the train

Convert an RGB-encoded image to grayscale labels. Args: arr_3d: int 2D image of color-coded labels on 3 channels palette: dict of

Convert an array of labels to RGB color-encoded image. Args: arr_2d: int 2D array of labels palette: dict of colors used (label n

Gets the dataset specified by name and return the related components. Args: dataset_name: string with the name of the dataset targ

fix the number of validation runs training will last for `epochs` or `max_steps`, whatever takes longer for small data_fraction and fixed batc

merge adjacent rasterio datasets into a single combined tile

Extract a fixed percentage of samples from an array of labels. Args: gt: a 2D array of int labels percentage: [0, 1] float Re

Sliding window generator over an input image. Args: image: 2D+ image to slide the window on, e.g. RGB or hyperspectral step: int

Test a model on a specific image

Compute inverse median frequency weights for class balancing. For each class i, it computes its frequency f_i, i.e the ratio between the num

Count the number of windows in an image. Args: image: 2D+ image to slide the window on, e.g. RGB or hyperspectral, ... step: int

Display the specified dataset. Args: img: 3D hyperspectral image gt: 2D array labels bands: tuple of RGB bands to select

downsample by keeping most frequent value in factorxfactor window

Plot sampled spectrums with mean + std for each class. Args: img: 3D hyperspectral image complete_gt: 2D array of labels

grid_size: int of the grid height and width return: pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/

For each spectral block in wavelengths, return the index of the closest spectral block in reference_wavelengths usecase: model pre-trained on

Browse an iterable by grouping n elements by n elements. Args: n: int, size of the groups iterable: the iterable to Browse Y

Compute and print metrics (accuracy, confusion matrix and F1 scores). Args: prediction: list of predicted labels target: list of

Plot the specified dictionary of spectrums. Args: spectrums: dictionary (name -> spectrum) of spectrums to plot vis: Visdom displ

tile image into multiple image_size,image_size patches and stack along batch dimension

Training loop to optimize a network for several epochs and a specified loss Args: net: a PyTorch model optimizer: a PyTorch

Ensure that boolean flags are correctly handeled