test_losses.py

import pytest
import torch

from basicsr.losses.basic_loss import CharbonnierLoss, L1Loss, MSELoss, WeightedTVLoss


@pytest.mark.parametrize('loss_class', [L1Loss, MSELoss, CharbonnierLoss])
def test_pixellosses(loss_class):
    """Test loss: pixel losses"""

    pred = torch.rand((1, 3, 4, 4), dtype=torch.float32)
    target = torch.rand((1, 3, 4, 4), dtype=torch.float32)
    loss = loss_class(loss_weight=1.0, reduction='mean')
    out = loss(pred, target, weight=None)
    assert isinstance(out, torch.Tensor)
    assert out.shape == torch.Size([])

    # -------------------- test with other reduction -------------------- #
    # reduction = none
    loss = loss_class(loss_weight=1.0, reduction='none')
    out = loss(pred, target, weight=None)
    assert isinstance(out, torch.Tensor)
    assert out.shape == (1, 3, 4, 4)
    # test with spatial weights
    weight = torch.rand((1, 3, 4, 4), dtype=torch.float32)
    out = loss(pred, target, weight=weight)
    assert isinstance(out, torch.Tensor)
    assert out.shape == (1, 3, 4, 4)

    # reduction = sum
    loss = loss_class(loss_weight=1.0, reduction='sum')
    out = loss(pred, target, weight=None)
    assert isinstance(out, torch.Tensor)
    assert out.shape == torch.Size([])

    # -------------------- test unsupported loss reduction -------------------- #
    with pytest.raises(ValueError):
        loss_class(loss_weight=1.0, reduction='unknown')


def test_weightedtvloss():
    """Test loss: WeightedTVLoss"""

    pred = torch.rand((1, 3, 4, 4), dtype=torch.float32)
    loss = WeightedTVLoss(loss_weight=1.0, reduction='mean')
    out = loss(pred, weight=None)
    assert isinstance(out, torch.Tensor)
    assert out.shape == torch.Size([])

    # test with spatial weights
    weight = torch.rand((1, 3, 4, 4), dtype=torch.float32)
    out = loss(pred, weight=weight)
    assert isinstance(out, torch.Tensor)
    assert out.shape == torch.Size([])

    # -------------------- test reduction = sum-------------------- #
    loss = WeightedTVLoss(loss_weight=1.0, reduction='sum')
    out = loss(pred, weight=None)
    assert isinstance(out, torch.Tensor)
    assert out.shape == torch.Size([])

    # test with spatial weights
    weight = torch.rand((1, 3, 4, 4), dtype=torch.float32)
    out = loss(pred, weight=weight)
    assert isinstance(out, torch.Tensor)
    assert out.shape == torch.Size([])

    # -------------------- test unsupported loss reduction -------------------- #
    with pytest.raises(ValueError):
        WeightedTVLoss(loss_weight=1.0, reduction='unknown')
    with pytest.raises(ValueError):
        WeightedTVLoss(loss_weight=1.0, reduction='none')