def test_image_clims_and_gamma(input_dtype, texture_format, num_channels,
clim_on_init, data_on_init):
"""Test image visual with clims and gamma on shader."""
size = (40, 40)
if texture_format == '__dtype__':
texture_format = input_dtype
shape = size + (num_channels, ) if num_channels > 0 else size
np.random.seed(0)
data = _make_test_data(shape, input_dtype)
orig_clim, new_clim = _get_orig_and_new_clims(input_dtype)
# 16-bit integers and above seem to have precision loss when scaled on the CPU
is_16int_cpu_scaled = (np.dtype(input_dtype).itemsize >= 2
and np.issubdtype(input_dtype, np.integer)
and texture_format is None)
clim_atol = 2 if is_16int_cpu_scaled else 1
gamma_atol = 3 if is_16int_cpu_scaled else 2
kwargs = {}
if clim_on_init:
kwargs['clim'] = orig_clim
if data_on_init:
kwargs['data'] = data
# default is RGBA, anything except auto requires reformat
set_data_fails = (num_channels != 4 and texture_format is not None
and texture_format != 'auto')
with TestingCanvas(size=size[::-1], bgcolor="w") as c:
image = Image(cmap='grays',
texture_format=texture_format,
parent=c.scene,
**kwargs)
if not data_on_init:
_set_image_data(image, data, set_data_fails)
if set_data_fails:
return
rendered = c.render()
_dtype = rendered.dtype
shape_ratio = rendered.shape[0] // data.shape[0]
rendered1 = downsample(rendered, shape_ratio,
axis=(0, 1)).astype(_dtype)
_compare_render(data, rendered1)
# adjust color limits
image.clim = new_clim
rendered2 = downsample(c.render(), shape_ratio,
axis=(0, 1)).astype(_dtype)
scaled_data = (np.clip(data, new_clim[0], new_clim[1]) -
new_clim[0]) / (new_clim[1] - new_clim[0])
_compare_render(scaled_data, rendered2, rendered1, atol=clim_atol)
# adjust gamma
image.gamma = 2
rendered3 = downsample(c.render(), shape_ratio,
axis=(0, 1)).astype(_dtype)
_compare_render(scaled_data**2, rendered3, rendered2, atol=gamma_atol)
def test_volume_clims_and_gamma():
"""Test volume visual with clims and gamma on shader.
currently just using np.ones since the angle of view made more complicated samples
challenging, but this confirms gamma and clims works in the shader.
"""
with TestingCanvas(size=(40, 40), bgcolor="k") as c:
v = c.central_widget.add_view(border_width=0, size=(40, 40))
data = np.ones((40, 40, 40)) / 2.5
volume = scene.visuals.Volume(
data,
clim=(0, 1),
interpolation='nearest',
parent=v.scene,
method='mip',
)
v.camera = 'arcball'
v.camera.fov = 0
v.camera.scale_factor = 40
v.camera.center = (19.5, 19.5, 19.5)
rendered = c.render()[..., 0]
_dtype = rendered.dtype
shape_ratio = rendered.shape[0] // data.shape[0]
rendered1 = downsample(rendered, shape_ratio,
axis=(0, 1)).astype(_dtype)
predicted = np.round(data.max(0) * 255).astype(np.uint8)
assert np.allclose(predicted, rendered1, atol=1)
# adjust contrast limits
new_clim = (0.3, 0.8)
volume.clim = new_clim
rendered2 = downsample(c.render()[..., 0], shape_ratio,
axis=(0, 1)).astype(_dtype)
scaled_data = np.clip((data - new_clim[0]) / np.diff(new_clim)[0], 0,
1)
predicted = np.round(scaled_data.max(0) * 255).astype(np.uint8)
assert np.allclose(predicted, rendered2)
assert not np.allclose(rendered1, rendered2, atol=10)
# adjust gamma
volume.gamma = 1.5
rendered3 = downsample(c.render()[..., 0], shape_ratio,
axis=(0, 1)).astype(_dtype)
predicted = np.round(
(scaled_data**volume.gamma).max(0) * 255).astype(np.uint8)
assert np.allclose(predicted, rendered3)
assert not np.allclose(rendered2, rendered3, atol=10)
def test_image_clims_and_gamma():
"""Test image visual with clims and gamma on shader."""
size = (40, 40)
with TestingCanvas(size=size, bgcolor="w") as c:
for three_d in (True,):
shape = size + ((3,) if three_d else ())
np.random.seed(0)
image = Image(cmap='grays', clim=[0, 1], parent=c.scene)
data = np.random.rand(*shape)
image.set_data(data)
rendered = c.render()
_dtype = rendered.dtype
shape_ratio = rendered.shape[0] // data.shape[0]
rendered1 = downsample(rendered, shape_ratio, axis=(0, 1)).astype(_dtype)
predicted = _make_rgba(data)
assert np.allclose(predicted, rendered1, atol=1)
# adjust contrast limits
new_clim = (0.3, 0.8)
image.clim = new_clim
rendered2 = downsample(c.render(), shape_ratio, axis=(0, 1)).astype(_dtype)
scaled_data = np.clip((data - new_clim[0]) / np.diff(new_clim)[0], 0, 1)
predicted = _make_rgba(scaled_data)
assert np.allclose(predicted, rendered2, atol=1)
assert not np.allclose(rendered1, rendered2, atol=10)
# adjust gamma
image.gamma = 2
rendered3 = downsample(c.render(), shape_ratio, axis=(0, 1)).astype(_dtype)
predicted = _make_rgba(scaled_data ** 2)
assert np.allclose(
predicted.astype(np.float), rendered3.astype(np.float), atol=2
)
assert not np.allclose(
rendered2.astype(np.float), rendered3.astype(np.float), atol=10
)
def test_volume_clims_and_gamma(texture_format, input_dtype, clim_on_init):
"""Test volume visual with clims and gamma on shader.
Test is parameterized based on ``texture_format`` and should produce
relatively the same results for each format.
Currently just using np.ones since the angle of view made more complicated samples
challenging, but this confirms gamma and clims works in the shader.
The VolumeVisual defaults to the "grays" colormap so although we compare
data using RGBA arrays, each R/G/B channel should be the same.
"""
size = (40, 40)
if texture_format == '__dtype__':
texture_format = input_dtype
np.random.seed(0) # make tests the same every time
data = _make_test_data(size[:1] * 3, input_dtype)
clim = (0, 1)
new_clim = (0.3, 0.8)
max = max_for_dtype(input_dtype)
if max != 1:
clim = (clim[0] * max, clim[1] * max)
new_clim = (new_clim[0] * max, new_clim[1] * max)
kwargs = {}
if clim_on_init:
kwargs['clim'] = clim
with TestingCanvas(size=size, bgcolor="k") as c:
v = c.central_widget.add_view(border_width=0)
volume = scene.visuals.Volume(data,
interpolation='nearest',
parent=v.scene,
method='mip',
texture_format=texture_format,
**kwargs)
v.camera = 'arcball'
v.camera.fov = 0
v.camera.scale_factor = 40.0
v.camera.center = (19.5, 19.5, 19.5)
rendered = c.render()
_dtype = rendered.dtype
shape_ratio = rendered.shape[0] // data.shape[0]
rendered1 = downsample(rendered, shape_ratio,
axis=(0, 1)).astype(_dtype)
predicted = data.max(axis=1)
compare_render(predicted, rendered1)
# adjust contrast limits
volume.clim = new_clim
rendered2 = downsample(c.render(), shape_ratio,
axis=(0, 1)).astype(_dtype)
scaled_data = (np.clip(data, new_clim[0], new_clim[1]) -
new_clim[0]) / (new_clim[1] - new_clim[0])
predicted = scaled_data.max(axis=1)
compare_render(predicted, rendered2, previous_render=rendered)
# adjust gamma
volume.gamma = 2
rendered3 = downsample(c.render(), shape_ratio,
axis=(0, 1)).astype(_dtype)
predicted = (scaled_data**2).max(axis=1)
compare_render(predicted, rendered3, previous_render=rendered2)
