| 12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879 |
- """ Layer/Module Helpers
- Hacked together by / Copyright 2020 Ross Wightman
- """
- from itertools import repeat
- import collections.abc
- # From PyTorch internals
- def _ntuple(n):
- """Return a function that converts input to an n-tuple.
- Scalar values are repeated n times, while iterables are converted to tuples.
- Strings are treated as scalars to avoid character-level splitting.
- Args:
- n: Target tuple length.
- Returns:
- Function that converts input to n-tuple.
- """
- def parse(x):
- if isinstance(x, collections.abc.Iterable) and not isinstance(x, str):
- return tuple(x)
- return tuple(repeat(x, n))
- return parse
- to_1tuple = _ntuple(1)
- to_2tuple = _ntuple(2)
- to_3tuple = _ntuple(3)
- to_4tuple = _ntuple(4)
- to_ntuple = _ntuple
- def make_divisible(v, divisor=8, min_value=None, round_limit=.9):
- """Adjust value to be divisible by a divisor, typically for channel counts.
- Rounds to the nearest multiple of divisor while ensuring the result doesn't
- fall below min_value or decrease by more than (1 - round_limit).
- Args:
- v: Value to adjust.
- divisor: Target divisor.
- min_value: Minimum acceptable value.
- round_limit: Prevent decrease beyond this fraction of original value.
- Returns:
- Adjusted value divisible by divisor.
- """
- min_value = min_value or divisor
- new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
- # Make sure that round down does not go down by more than 10%.
- if new_v < round_limit * v:
- new_v += divisor
- return new_v
- def extend_tuple(x, n):
- """Pad a tuple to length n by repeating the last value.
- If input is shorter than n, extends by repeating the last element.
- If input is longer than n, truncates to n.
- Args:
- x: Input value, tuple, or list.
- n: Target length.
- Returns:
- Tuple of length n.
- """
- if not isinstance(x, (tuple, list)):
- x = (x,)
- else:
- x = tuple(x)
- pad_n = n - len(x)
- if pad_n <= 0:
- return x[:n]
- return x + (x[-1],) * pad_n
|