File size: 1,886 Bytes
17cd746 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 |
# Copyright (c) 2023-2024, Zexin He
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch
import torch.nn as nn
from einops import rearrange
__all__ = ['PixelLoss']
class PixelLoss(nn.Module):
"""
Pixel-wise loss between two images.
"""
def __init__(self, option: str = 'mse'):
super().__init__()
self.loss_fn = self._build_from_option(option)
@staticmethod
def _build_from_option(option: str, reduction: str = 'none'):
if option == 'mse':
return nn.MSELoss(reduction=reduction)
elif option == 'l1':
return nn.L1Loss(reduction=reduction)
else:
raise NotImplementedError(f'Unknown pixel loss option: {option}')
@torch.compile
def forward(self, x, y, conf_sigma=None, only_sym_conf=False):
"""
Assume images are channel first.
Args:
x: [N, M, C, H, W]
y: [N, M, C, H, W]
Returns:
Mean-reduced pixel loss across batch.
"""
N, M, C, H, W = x.shape
x = rearrange(x, "n m c h w -> (n m) c h w")
y = rearrange(y, "n m c h w -> (n m) c h w")
image_loss = self.loss_fn(x, y)
image_loss = image_loss.mean(dim=[1, 2, 3])
batch_loss = image_loss.reshape(N, M).mean(dim=1)
all_loss = batch_loss.mean()
return all_loss
|