class
ChannelDropout
extends
PhotometricTransform[ChannelDropParams]ChannelDropout(channel_drop_range: tuple[int, int] = (1, 1), fill_value: float = 0.0, p: float = 0.5)Replace random channels with a constant fill value (Albumentations ChannelDropout).
Samples a count n uniformly from channel_drop_range and
selects n distinct channels without replacement. Each chosen
channel is replaced by fill_value via a multiplicative keep-mask
- additive fill, so the op stays autograd-friendly.
Parameters
channel_drop_range(int, int)= (1, 1)Inclusive range from which the number of channels to drop is
sampled (
randint(lo, hi + 1)).fill_valuefloat= 0.0Value written into the dropped channels.
pfloat= 0.5Probability of applying the transform.
Examples
>>> import lucid
>>> import lucid.utils.transforms as T
>>> tf = T.ChannelDropout(channel_drop_range=(1, 1), fill_value=0.0, p=1.0)
>>> out = tf(T.Image(lucid.rand(3, 32, 32))).data
>>> tuple(out.shape)
(3, 32, 32)Used by 1
Constructors
1Instance methods
1Sample per-call random parameters for ChannelDropout.
Parameters
imgTensorImage tensor; its channel-axis size is read so that the
sample-without-replacement loop never returns duplicate or
out-of-range indices.
Returns
ChannelDropParamsCarries channels — a tuple of distinct channel indices
(sampled without replacement) to replace with
fill_value in the apply step.
Notes
The number of dropped channels is drawn from
channel_drop_range; it is clamped to the available channel
count if the high bound exceeds C.