tools/math_utils.py

import math
import torch


## [-1,1]
def tensor2log(x):
    a = (math.e - 1.) / 2.
    b = (math.e + 1.) / 2.
    x = a * x + b
    return torch.log(x).float()


def log2tensor(x):
    a = 2. / (math.e - 1.)
    b = (math.e + 1.) / (1. - math.e)
    x = torch.exp(x)
    x = a * x + b
    return x.float()


## [0,1]
def _tensor2log(x):
    a = math.e - 1.
    b = 1.
    x = a * x + b
    return torch.log(x).float()


def _log2tensor(x):
    a = 1. / (math.e - 1.)
    b = -a
    x = torch.exp(x)
    x = a * x + b
    return x.float()


if __name__ == '__main__':
    inputx = torch.rand(1, 3, 64, 64)
    print(torch.min(inputx), torch.max(inputx))

    out = _tensor2log(inputx)
    print(torch.min(out), torch.max(out))

    out = _log2tensor(out)
    print(torch.min(out), torch.max(out))
    print(torch.mean(out - inputx))