cutcutcodec.core.analysis.video.metric.lpips_torch

Compute a differenciable batched torch lpips.

Functions

lpips_torch(im1, im2[, net, threads])

Pure torch implementation of cutcutcodec.core.analysis.video.metric.lpips().

Details

cutcutcodec.core.analysis.video.metric.lpips_torch.lpips_torch(im1: Tensor, im2: Tensor, net: str | Module = 'alex', threads: int = 0) → Tensor

Pure torch implementation of cutcutcodec.core.analysis.video.metric.lpips().

Examples

>>> import torch
>>> from cutcutcodec.core.analysis.video.metric.lpips_torch import lpips_torch
>>> _ = torch.manual_seed(0)
>>> im1 = torch.rand(2, 4, 720, 1080, 3)
>>> im2 = 0.8 * im1 + 0.2 * torch.rand(2, 4, 720, 1080, 3)
>>> lpips_torch(im1[0, 0], im1[0, 0])
tensor(0.)
>>> lpips_torch(im1, im2)
tensor([[0.0492, 0.0443, 0.0468, 0.0469],
        [0.0444, 0.0445, 0.0470, 0.0446]])
>>>