kaishi.image.labelers.generic_convnet

Class definition for generic convnet labeler.

Module Contents

class kaishi.image.labelers.generic_convnet.LabelerGenericConvnet

Bases: kaishi.core.pipeline_component.PipelineComponent

Use pre-trained ConvNet to predict image labels (e.g. stretched, rotated, etc.).

This labeler uses a default configured kaishi.image.model.Model object where the output layer is assumed to have 6 values ranging 0 to 1, where the labels are [DOCUMENT, RECTIFIED, ROTATED_RIGHT, ROTATED_LEFT, UPSIDE_DOWN, STRETCHED].

__call__(self, dataset)

Perform the labeling operation on an image dataset.

Parameters

dataset (kaishi.image.dataset.ImageDataset) – kaishi image dataset