COSIE.image_preprocessing

Functions

`combine_mask`	Modify a grayscale image by applying a binary mask.
`create_model`	Create and load a pre-trained Vision Transformer (ViT-L/16) model from a HuggingFace-compatible checkpoint.
`extract_features`	Extracts both global and local visual features from an input image batch using a pretrained model.
`generate_homograph`	Generate a geometric transformation matrix from source to destination keypoints, using either rigid or affine transformation.
`generate_pxl_location_from_mask`	Extract pixel and spatial coordinates of 16×16 superpixels whose top-left pixels fall within the white (255) region of a binary mask.
`get_white_superpixel_centers`	Identify the center coordinates of superpixels that are entirely white (255) in a binary image.
`image_feature_extraction`	Extract image features at cell locations using a pretrained Vision Transformer (ViT) model and save the output to disk as a pickle file.
`load_image`	Efficiently load an image file and convert it to a NumPy array.
`load_pickle`	Load a Python object from a .pkl file.
`mkdir`	Create the parent directory for a given file or directory path if it does not already exist.
`rescale_image`	Rescale an image by a given scale factor using skimage.transform.rescale.
`save_jpg`	Convert a NumPy array into a JPEG image and save it to disk.
`save_pickle`	Save a Python object to a .pkl file using the pickle module.
`transform_coordinates`	Apply a geometric transformation matrix (e.g., from generate_homograph) to 2D coordinates.

Classes

A PyTorch-compatible dataset for extracting 224×224 image patches centered at specified cell coordinates from a high-resolution RGB image.