Skip to content

🐍 Geometric Computer Vision Library for Spatial AI

14.3 KB
Raw Blame
# LICENSE HEADER MANAGED BY add-license-header
#
# Copyright 2018 Kornia Team
#
# 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
#
#     http://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 pytest
import torch

import kornia

from testing.base import (
    BaseTester,
    assert_close,  # test utils
)


class TestRawToRgb(BaseTester):
    def test_smoke(self, device, dtype):
        C, H, W = 1, 4, 6
        img = torch.rand(C, H, W, device=device, dtype=dtype)
        assert isinstance(kornia.color.raw_to_rgb(img, kornia.color.CFA.BG), torch.Tensor)

    @pytest.mark.parametrize("batch_size, height, width", [(1, 6, 4), (2, 2, 4), (3, 4, 2)])
    def test_cardinality(self, device, dtype, batch_size, height, width):
        img = torch.ones(batch_size, 1, height, width, device=device, dtype=dtype)
        assert kornia.color.raw_to_rgb(img, kornia.color.CFA.BG).shape == (batch_size, 3, height, width)

    def test_exception(self, device, dtype):
        with pytest.raises(TypeError):
            assert kornia.color.raw_to_rgb([0.0], kornia.color.CFA.BG)

        img = torch.ones(1, 1, device=device, dtype=dtype)
        with pytest.raises(ValueError):
            assert kornia.color.raw_to_rgb(img, kornia.color.CFA.GB)

        img = torch.ones(2, 1, 1, device=device, dtype=dtype)
        with pytest.raises(ValueError):
            assert kornia.color.raw_to_rgb(img, kornia.color.CFA.RG)

        img = torch.ones(1, 3, 1, 1, device=device, dtype=dtype)
        with pytest.raises(ValueError):
            assert kornia.color.raw_to_rgb(img, kornia.color.CFA.GR)

        # dimensionality test
        img = torch.ones(3, 2, 1, device=device, dtype=dtype)
        with pytest.raises(ValueError):
            assert kornia.color.raw_to_rgb(img, kornia.color.CFA.GR)

        # dimensionality test
        img = torch.ones(3, 1, 2, device=device, dtype=dtype)
        with pytest.raises(ValueError):
            assert kornia.color.raw_to_rgb(img, kornia.color.CFA.GR)

    @pytest.mark.parametrize("cfa", list(kornia.color.CFA))
    def test_transposed_batch(self, device, dtype, cfa):
        image = torch.rand(2, 3, 1, 4, 6, device=device, dtype=dtype).transpose(0, 1)
        assert not image.is_contiguous()
        expected = torch.stack(
            [torch.stack([kornia.color.raw_to_rgb(frame, cfa) for frame in batch]) for batch in image]
        )
        actual = kornia.color.raw_to_rgb(image, cfa)
        self.assert_close(actual, expected)
        self.assert_close(kornia.color.rgb_to_raw(actual, cfa), image)

    def test_transposed_batch_gradcheck(self, device):
        image = torch.rand(2, 2, 1, 2, 2, device=device, dtype=torch.float64, requires_grad=True)
        self.gradcheck(lambda x: kornia.color.raw_to_rgb(x.transpose(0, 1), kornia.color.CFA.BG), (image,))

    def test_transposed_batch_dynamo(self, device, dtype, torch_optimizer):
        image = torch.rand(2, 3, 1, 4, 6, device=device, dtype=dtype).transpose(0, 1)
        op = kornia.color.raw_to_rgb
        self.assert_close(torch_optimizer(op)(image, kornia.color.CFA.BG), op(image, kornia.color.CFA.BG))

    # With he current implementations we should get back an identical raw representation when doing raw -> rgb -> raw
    # Note that with more advanced implementations this may not necessarily be true or desirable
    def test_forth_and_back(self, device, dtype):  # skipcq: PYL-R0201
        data = torch.rand(1, 80, 80, device=device, dtype=dtype)
        raw = kornia.color.rgb_to_raw
        rgb = kornia.color.raw_to_rgb

        for x in kornia.color.CFA:
            data_out = raw(rgb(data, cfa=x), cfa=x)
            assert_close(data_out, data)

    # make sure different cfas are actually different
    def test_cfas_not_the_same(self, device, dtype):  # skipcq: PYL-R0201
        data = torch.rand(1, 16, 16, device=device, dtype=dtype)
        assert (
            torch.max(
                kornia.color.raw_to_rgb(data, kornia.color.CFA.BG) - kornia.color.raw_to_rgb(data, kornia.color.CFA.RG)
            )
            > 0.0
        )

    # The outcome will be very different for different implementations
    # Here we compare against a current baseline, it is safe to update this if the underlying algorithm changes
    def test_functional(self, device, dtype):  # skipcq: PYL-R0201
        data = torch.tensor(
            [[[1, 0.5, 0.2, 0.4], [0.75, 0.25, 0.8, 0.3], [0.65, 0.15, 0.7, 0.2], [0.55, 0.5, 0.6, 0.1]]],
            device=device,
            dtype=dtype,
        )
        # checked by hand as correct interpolation. Note the ugly replication that happens for Red on the last column
        # and row. We shall accept to live with that
        expected = torch.tensor(
            [
                [
                    [1.0000, 0.6000, 0.2000, 0.2000],
                    [0.8250, 0.6375, 0.4500, 0.4500],
                    [0.6500, 0.6750, 0.7000, 0.7000],
                    [0.6500, 0.6750, 0.7000, 0.7000],
                ],
                [
                    [0.6250, 0.5000, 0.6250, 0.4000],
                    [0.7500, 0.5500, 0.8000, 0.5500],
                    [0.4000, 0.1500, 0.4375, 0.2000],
                    [0.5500, 0.3625, 0.6000, 0.4000],
                ],
                [
                    [0.2500, 0.2500, 0.2750, 0.3000],
                    [0.2500, 0.2500, 0.2750, 0.3000],
                    [0.3750, 0.3750, 0.2875, 0.2000],
                    [0.5000, 0.5000, 0.3000, 0.1000],
                ],
            ],
            device=device,
            dtype=dtype,
        )

        img_rgb = kornia.color.raw_to_rgb(data, kornia.color.raw.CFA.BG)
        assert_close(img_rgb, expected)

    # If we roll the data and the different CFAs they give the same result (expect on edges!)
    def test_cfa_on_rolled(self, device, dtype):  # skipcq: PYL-R0201
        data = torch.rand(1, 1, 8, 8, device=device, dtype=dtype)
        bgres = kornia.color.raw_to_rgb(data, kornia.color.raw.CFA.BG)
        gbres = kornia.color.raw_to_rgb(data.roll((0, 1), (-2, -1)), kornia.color.raw.CFA.GB)
        grres = kornia.color.raw_to_rgb(data.roll((1, 0), (-2, -1)), kornia.color.raw.CFA.GR)
        rgres = kornia.color.raw_to_rgb(data.roll((1, 1), (-2, -1)), kornia.color.raw.CFA.RG)

        assert_close(bgres[:, :, 1:5, 1:5], gbres[:, :, 1:5, 2:6])
        assert_close(bgres[:, :, 1:5, 1:5], grres[:, :, 2:6, 1:5])
        assert_close(bgres[:, :, 1:5, 1:5], rgres[:, :, 2:6, 2:6])

    def test_gradcheck(self, device, dtype):
        B, C, H, W = 2, 1, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=torch.float64, requires_grad=True)
        assert self.gradcheck(kornia.color.raw_to_rgb, (img, kornia.color.raw.CFA.BG))

    def test_jit(self, device, dtype):
        B, C, H, W = 2, 1, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        op = kornia.color.raw_to_rgb
        op_jit = torch.jit.script(op)
        assert_close(op(img, kornia.color.raw.CFA.BG), op_jit(img, kornia.color.raw.CFA.BG))

    def test_module(self, device, dtype):
        B, C, H, W = 2, 1, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        raw_ops = kornia.color.RawToRgb(kornia.color.raw.CFA.BG).to(device, dtype)
        raw_fcn = kornia.color.raw_to_rgb
        assert_close(raw_ops(img), raw_fcn(img, kornia.color.raw.CFA.BG))


class TestRgbToRaw(BaseTester):
    def test_smoke(self, device, dtype):
        C, H, W = 3, 4, 6
        img = torch.rand(C, H, W, device=device, dtype=dtype)
        assert isinstance(kornia.color.rgb_to_raw(img, kornia.color.raw.CFA.BG), torch.Tensor)

    @pytest.mark.parametrize("batch_size, height, width", [(1, 3, 4), (2, 2, 4), (3, 4, 1)])
    def test_cardinality(self, device, dtype, batch_size, height, width):
        img = torch.ones(batch_size, 3, height, width, device=device, dtype=dtype)
        assert kornia.color.rgb_to_raw(img, kornia.color.raw.CFA.GR).shape == (batch_size, 1, height, width)

    def test_exception(self, device, dtype):
        with pytest.raises(TypeError):
            assert kornia.color.rgb_to_raw([0.0], kornia.color.raw.CFA.RG)

        img = torch.ones(1, 1, device=device, dtype=dtype)
        with pytest.raises(ValueError):
            assert kornia.color.rgb_to_raw(img, kornia.color.raw.CFA.BG)

        img = torch.rand(1, 3, 2, 3, device=device, dtype=dtype)
        with pytest.raises(ValueError, match="Unsupported CFA value"):
            kornia.color.rgb_to_raw(img, "bad")

        # Reverse test in rawtorgb is sufficient functional test

    def test_gradcheck(self, device, dtype):
        B, C, H, W = 2, 3, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=torch.float64, requires_grad=True)
        assert self.gradcheck(kornia.color.rgb_to_raw, (img, kornia.color.raw.CFA.BG))

    def test_jit(self, device, dtype):
        B, C, H, W = 2, 3, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        op = kornia.color.rgb_to_raw
        op_jit = torch.jit.script(op)
        assert_close(op(img, kornia.color.raw.CFA.BG), op_jit(img, kornia.color.raw.CFA.BG))

    def test_module(self, device, dtype):
        B, C, H, W = 2, 3, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        raw_ops = kornia.color.RgbToRaw(kornia.color.raw.CFA.BG).to(device, dtype)
        raw_fcn = kornia.color.rgb_to_raw
        assert_close(raw_ops(img), raw_fcn(img, kornia.color.raw.CFA.BG))


class TestRawToRgb2x2Downscaled(BaseTester):
    def test_smoke(self, device, dtype) -> None:
        C, H, W = 1, 4, 6
        img = torch.rand(C, H, W, device=device, dtype=dtype)
        assert isinstance(kornia.color.raw_to_rgb_2x2_downscaled(img, kornia.color.CFA.BG), torch.Tensor)

    @pytest.mark.parametrize(
        "batch_size, height, width",
        [((1,), 6, 4), ((), 6, 4), ((), 0, 0), ((4, 5), 0, 0), ((2,), 2, 4), ((3,), 4, 2), ((4, 2, 1), 4, 2)],
    )
    def test_cardinality(self, device, dtype, batch_size, height, width):
        img = torch.ones(*batch_size, 1, height, width, device=device, dtype=dtype)
        assert kornia.color.raw_to_rgb_2x2_downscaled(img, kornia.color.CFA.BG).shape == (
            *batch_size,
            3,
            height // 2,
            width // 2,
        )

    def test_exception(self, device, dtype):
        from kornia.core.exceptions import BaseError, ShapeError

        with pytest.raises(BaseError) as errinf:
            kornia.color.raw_to_rgb_2x2_downscaled([0.0], kornia.color.CFA.BG)
        assert "Input type is not a torch.Tensor" in str(errinf.value)

        img = torch.ones(1, 1, device=device, dtype=dtype)
        with pytest.raises(ShapeError) as errinf:
            kornia.color.raw_to_rgb_2x2_downscaled(img, kornia.color.CFA.GB)
        assert "Shape dimension mismatch" in str(errinf.value) or "Expected shape" in str(errinf.value)

        img = torch.ones(2, 2, 2, device=device, dtype=dtype)
        with pytest.raises(ShapeError) as errinf:
            kornia.color.raw_to_rgb_2x2_downscaled(img, kornia.color.CFA.RG)
        assert "Shape dimension mismatch" in str(errinf.value) or "Expected shape" in str(errinf.value)

        img = torch.ones(1, 3, 2, device=device, dtype=dtype)
        with pytest.raises(Exception) as errinf:
            kornia.color.raw_to_rgb_2x2_downscaled(img, kornia.color.CFA.GR)
        assert "Input H&W must be evenly divisible by 2. Got" in str(errinf)

        img = torch.ones(1, 2, 3, device=device, dtype=dtype)
        with pytest.raises(Exception) as errinf:
            kornia.color.raw_to_rgb_2x2_downscaled(img, kornia.color.CFA.GR)
        assert "Input H&W must be evenly divisible by 2. Got" in str(errinf)

        img = torch.ones(1, 4, 8, device=device, dtype=dtype)
        nonexistent_cfa = 195162495283
        with pytest.raises(ValueError) as errinf:
            kornia.color.raw_to_rgb_2x2_downscaled(img, nonexistent_cfa)
        assert "Unsupported CFA Got" in str(errinf)

    @pytest.mark.parametrize(
        "cfa, expected_rgb",
        [
            (kornia.color.raw.CFA.BG, [[[0.12, 0.46]], [[0.33, 0.505]], [[0.00, 0.89]]]),
            (kornia.color.raw.CFA.GB, [[[0.43, 0.58]], [[0.06, 0.675]], [[0.23, 0.43]]]),
            (kornia.color.raw.CFA.RG, [[[0.0, 0.89]], [[0.33, 0.505]], [[0.12, 0.46]]]),
            (kornia.color.raw.CFA.GR, [[[0.23, 0.43]], [[0.06, 0.675]], [[0.43, 0.58]]]),
        ],
    )
    def test_functional(self, device, dtype, cfa, expected_rgb):
        data = torch.tensor([[[0.12, 0.43, 0.46, 0.58], [0.23, 0.00, 0.43, 0.89]]], device=device, dtype=dtype)
        expected_rgb = torch.tensor(expected_rgb, device=device, dtype=dtype)

        img_rgb = kornia.color.raw_to_rgb_2x2_downscaled(data, cfa)
        self.assert_close(img_rgb, expected_rgb)

    def test_gradcheck(self, device, dtype):
        B, C, H, W = 2, 1, 8, 8
        img = torch.ones(B, C, H, W, device=device, dtype=torch.float64, requires_grad=True)
        self.gradcheck(
            kornia.color.raw_to_rgb_2x2_downscaled, (img, kornia.color.raw.CFA.BG), raise_exception=True, fast_mode=True
        )

    def test_jit(self, device, dtype):
        B, C, H, W = 2, 1, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        op = kornia.color.raw_to_rgb_2x2_downscaled
        op_jit = torch.jit.script(op)
        self.assert_close(op(img, kornia.color.raw.CFA.BG), op_jit(img, kornia.color.raw.CFA.BG))

    def test_module(self, device, dtype):
        B, C, H, W = 2, 1, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        raw_ops = kornia.color.RawToRgb2x2Downscaled(kornia.color.raw.CFA.BG).to(device, dtype)
        raw_fcn = kornia.color.raw_to_rgb_2x2_downscaled
        self.assert_close(raw_ops(img), raw_fcn(img, kornia.color.raw.CFA.BG))

    def test_dynamo(self, device, dtype, torch_optimizer):
        B, C, H, W = 2, 1, 4, 4
        img = torch.ones(B, C, H, W, device=device, dtype=dtype)
        op = kornia.color.raw_to_rgb_2x2_downscaled
        op_optimized = torch_optimizer(op)
        self.assert_close(op(img, kornia.color.CFA.BG), op_optimized(img, kornia.color.CFA.BG))