TextRecogGeneralAug

class mmocr.datasets.transforms.TextRecogGeneralAug

A general geometric augmentation tool for text images in the CVPR 2020 paper “Learn to Augment: Joint Data Augmentation and Network Optimization for Text Recognition”. It applies distortion, stretching, and perspective transforms to an image.

This implementation is adapted from https://github.com/RubanSeven/Text-Image-Augmentation-python/blob/master/augment.py # noqa

TODO: Split this transform into three transforms.

Required Keys:

• img

Modified Keys:

• img

• img_shape

tia_distort(img, segment=4)

Image distortion.

Parameters

• img (np.ndarray) – The image.

• segment (int) – The number of segments to divide the image along the width. Defaults to 4.

Return type

numpy.ndarray

tia_perspective(img)

Image perspective transformation.

Parameters

• img (np.ndarray) – The image.

• segment (int) – The number of segments to divide the image along the width. Defaults to 4.

Return type

numpy.ndarray

tia_stretch(img, segment=4)

Image stretching.

Parameters

• img (np.ndarray) – The image.

• segment (int) – The number of segments to divide the image along the width. Defaults to 4.

Return type

numpy.ndarray

transform(results)

Call function to pad images.

Parameters

results (dict) – Result dict from loading pipeline.

Returns

Updated result dict.

Return type

dict

warp_mls(src, src_pts, dst_pts, dst_w, dst_h, trans_ratio=1.0)

Warp the image.

Parameters

• src (numpy.ndarray) –

• src_pts (List[int]) –

• dst_pts (List[int]) –

• dst_w (int) –

• dst_h (int) –

• trans_ratio (float) –

Return type

numpy.ndarray