RandomGrayscale

class mmpretrain.datasets.transforms.RandomGrayscale(prob=0.1, keep_channels=False, channel_weights=(1.0, 1.0, 1.0), color_format='bgr')

Randomly convert image to grayscale with a probability.

Required Key:

• img

Modified Key:

• img

Added Keys:

• grayscale

• grayscale_weights

Parameters:

• prob (float) – Probability that image should be converted to grayscale. Defaults to 0.1.

• keep_channels (bool) – Whether keep channel number the same as input. Defaults to False.

• channel_weights (tuple) – The grayscale weights of each channel, and the weights will be normalized. For example, (1, 2, 1) will be normalized as (0.25, 0.5, 0.25). Defaults to (1., 1., 1.).

• color_format (str) – Color format set to be any of ‘bgr’, ‘rgb’, ‘hsv’. Note: ‘hsv’ image will be transformed into ‘bgr’ format no matter whether it is grayscaled. Defaults to ‘bgr’.

transform(results)

Apply random grayscale on results.

Parameters:

results (dict) – Result dict contains the data to transform.

Returns:

Results with grayscale image.

Return type:

dict