Skip to content

Core

Core vision functionality for computer vision tasks.

This module provides specialized learners and utilities for computer vision tasks. It extends the base Learner class with vision-specific functionality such as batch visualization and image processing capabilities.

Classes:

  • VisionLearner : Learner –

    Learner specialized for computer vision tasks with batch visualization.

Examples:

>>> # Create a vision learner
>>> learner = VisionLearner(model, dls, loss_func, opt_func)
>>>
>>> # Show a batch of images
>>> learner.show_batch(sample_sz=4, figsize=(12, 8))

VisionLearner

Bases: Learner

Learner specialized for computer vision tasks.

This class extends the base Learner with vision-specific functionality, particularly for visualizing batches of images during training and inference. It provides convenient methods for displaying image data and monitoring training progress in computer vision applications.

Examples:

>>> # Create a vision learner with a CNN model
>>> model = torchvision.models.resnet18(pretrained=True)
>>> learner = VisionLearner(model, dls, loss_func, opt_func)
>>>
>>> # Show a batch of images
>>> learner.show_batch(sample_sz=4)
>>>
>>> # Train the model
>>> learner.fit_one_cycle(10)

show_batch(sample_sz=1, callbacks=None, **kwargs)

Show a batch of images for visualization.

This method runs a single forward pass through the model and displays the input images. It's useful for inspecting the data being fed to the model and verifying data preprocessing.

Parameters:

  • sample_sz (int, default: 1 ) –

    Number of input samples to show from the batch.

  • callbacks (Iterable[Callback] | None, default: None ) –

    Additional callbacks to add temporarily for this visualization. These callbacks will be removed after the method completes.

  • **kwargs –

    Additional keyword arguments passed to show_images function. Common options include: - figsize : tuple, default=(10, 10) Figure size in inches (width, height) - nrows : int, default=None Number of rows in the grid - ncols : int, default=None Number of columns in the grid - title : str, default=None Title for the figure - cmap : str, default=None Colormap for grayscale images

Notes

This method temporarily adds a SingleBatchForwardCallback to ensure only one batch is processed, regardless of the current training state. The method will automatically clean up any additional callbacks that were passed in.

Examples:

>>> # Show 4 images from the current batch
>>> learner.show_batch(sample_sz=4, figsize=(12, 8))
>>>
>>> # Show images with custom grid layout
>>> learner.show_batch(sample_sz=9, nrows=3, ncols=3)
>>>
>>> # Show images with a title
>>> learner.show_batch(sample_sz=2, title="Training Batch")