pytorch-image

Minimalistic library designed to be an interface for the model input with a lightweight Image class.

from my_model import MyModel, Image

image = Image.from_numpy(numpy_image)
predictions = model(image)

It also offers a simple interface for essential image operations such as loading, resizing, augmenting, and saving.

Installation

pip install pytorch-image

poetry add pytorch-image

Usage

from pytorch_image import Image
import albumentations


image = Image.open("path/to/image.png")

image.augment(albumentations.HorizontalFlip())

image.torch()

Available Methods

The Image class provides the following methods:

Class Methods

• Image.open(path): Open an image file and create an Image instance.
• Image.from_pil_image(pil_image): Create an Image instance from a PIL Image.
• Image.from_numpy(numpy_image): Create an Image instance from a NumPy array. Expects HWC format, uint8 (0-255).
• Image.from_torch(torch_tensor): Create an Image instance from a PyTorch tensor. Expects NCHW format, float (0-1).
• Image.cat(images, dim=0): Concatenate multiple Image instances.

Instance Methods

• resize(shape, mode="bicubic", align_corners=None, antialias=None): Resize the image.
• augment(augmenter): Apply augmentations to the image.
• map(func): Apply a function to the image data.
• replace(**kwargs): Create a new Image instance with replaced attributes.
• torch(): Get the image data as a PyTorch tensor.
• numpy(): Get the image data as a NumPy array.
• representation(): Get a normalized representation of the image data.
• save(path): Save the image to a file.
• pil_image(): Convert the image to a PIL Image.

Properties and Special Methods

• shape: Get the shape of the image data.
• __len__(): Get the number of images in the batch.
• __iter__(): Iterate over images in the batch.
• __getitem__(index): Get a specific image from the batch.
• _repr_png_(): Get a PNG representation for Jupyter notebook display.

For detailed information on each method, please refer to the docstrings in the source code.