def test_relabel_consistency(n_rays, eps, plot=False):
""" test whether an already star-convex label image gets perfectly relabeld"""
# img = random_image((128, 123))
lbl1 = circle_image(shape=(32, 32), radius=8, eps=eps)
lbl1 = relabel_image_stardist(lbl1, n_rays)
lbl2 = relabel_image_stardist(lbl1, n_rays)
rel_error = 1 - np.count_nonzero(np.bitwise_and(
lbl1 > 0, lbl2 > 0)) / np.count_nonzero(lbl1 > 0)
print(rel_error)
assert rel_error < 1e-1
if plot:
import matplotlib.pyplot as plt
plt.figure(num=1, figsize=(8, 4))
plt.subplot(1, 3, 1)
plt.imshow(lbl1)
plt.title("GT")
plt.subplot(1, 3, 2)
plt.imshow(lbl2)
plt.title("Reco")
plt.subplot(1, 3, 3)
plt.imshow(1 * (lbl1 > 0) + 2 * (lbl2 > 0))
plt.title("Overlay")
plt.tight_layout()
plt.show()
return lbl1, lbl2
def test_model(tmpdir, n_rays, grid, n_channel, backbone):
img = circle_image(shape=(64, 80, 96))
imgs = np.repeat(img[np.newaxis], 3, axis=0)
if n_channel is not None:
imgs = np.repeat(imgs[..., np.newaxis], n_channel, axis=-1)
else:
n_channel = 1
X = imgs + .6 * np.random.uniform(0, 1, imgs.shape)
Y = (imgs if imgs.ndim == 4 else imgs[..., 0]).astype(int)
conf = Config3D(
backbone=backbone,
rays=n_rays,
grid=grid,
n_channel_in=n_channel,
use_gpu=False,
train_epochs=1,
train_steps_per_epoch=2,
train_batch_size=2,
train_loss_weights=(4, 1),
train_patch_size=(48, 64, 64),
)
model = StarDist3D(conf, name='stardist', basedir=str(tmpdir))
model.train(X, Y, validation_data=(X[:2], Y[:2]))
ref = model.predict(X[0])
res = model.predict(X[0],
n_tiles=((1, 2, 3) if X[0].ndim == 3 else
(1, 2, 3, 1)))
def test_model(tmpdir, n_rays, grid, n_channel):
img = circle_image(shape=(160,160))
imgs = np.repeat(img[np.newaxis], 3, axis=0)
if n_channel is not None:
imgs = np.repeat(imgs[...,np.newaxis], n_channel, axis=-1)
else:
n_channel = 1
X = imgs+.6*np.random.uniform(0,1,imgs.shape)
Y = (imgs if imgs.ndim==3 else imgs[...,0]).astype(int)
conf = Config2D (
n_rays = n_rays,
grid = grid,
n_channel_in = n_channel,
use_gpu = False,
train_epochs = 1,
train_steps_per_epoch = 2,
train_batch_size = 2,
train_loss_weights = (4,1),
train_patch_size = (128,128),
)
model = StarDist2D(conf, name='stardist', basedir=str(tmpdir))
model.train(X, Y, validation_data=(X[:2],Y[:2]))
def test_relabel_consistency(n_rays, eps, plot = False):
""" test whether an already star-convex label image gets perfectly relabeld"""
rays = Rays_GoldenSpiral(n_rays, anisotropy = 1./np.array(eps))
# img = random_image((128, 123))
lbl1 = circle_image(shape=(32,32,32), radius=8, eps = eps)
lbl1 = relabel_image_stardist3D(lbl1, rays)
lbl2 = relabel_image_stardist3D(lbl1, rays)
rel_error = 1-np.count_nonzero(np.bitwise_and(lbl1>0, lbl2>0))/np.count_nonzero(lbl1>0)
print(rel_error)
assert rel_error<1e-1
if plot:
import matplotlib.pyplot as plt
plt.figure(num=1, figsize=(8,4))
plt.subplot(2,3,1);plt.imshow(np.max(lbl1,0));plt.title("GT")
plt.subplot(2,3,2);plt.imshow(np.max(lbl2,0));plt.title("Reco")
plt.subplot(2,3,3);plt.imshow(np.max(1*(lbl1>0)+2*(lbl2>0),0));plt.title("Overlay")
plt.subplot(2,3,4);plt.imshow(np.max(lbl1,1));plt.title("GT")
plt.subplot(2,3,5);plt.imshow(np.max(lbl2,1));plt.title("Reco")
plt.subplot(2,3,6);plt.imshow(np.max(1*(lbl1>0)+2*(lbl2>0),1));plt.title("Overlay")
plt.tight_layout()
plt.show()
return lbl1, lbl2
def test_grid(grid,shape):
ss_grid = tuple(slice(0, None, g) for g in grid)
lbl = circle_image(shape=shape, radius=min(shape)//3)
n_rays = 32
d1 = star_dist(lbl, n_rays=n_rays)
d1 = d1[ss_grid]
d2 = star_dist(lbl, n_rays=n_rays, grid=grid)
assert np.allclose(d1,d2)
return lbl, d1, d2
def test_model(tmpdir, n_rays, grid, n_channel, backbone):
img = circle_image(shape=(64, 80, 96))
imgs = np.repeat(img[np.newaxis], 3, axis=0)
if n_channel is not None:
imgs = np.repeat(imgs[..., np.newaxis], n_channel, axis=-1)
else:
n_channel = 1
X = imgs + .6 * np.random.uniform(0, 1, imgs.shape)
Y = (imgs if imgs.ndim == 4 else imgs[..., 0]).astype(int)
conf = Config3D(
backbone=backbone,
rays=n_rays,
grid=grid,
n_channel_in=n_channel,
use_gpu=False,
train_epochs=1,
train_steps_per_epoch=2,
train_batch_size=2,
train_loss_weights=(4, 1),
train_patch_size=(48, 64, 64),
)
model = StarDist3D(conf, name='stardist', basedir=str(tmpdir))
model.train(X, Y, validation_data=(X[:2], Y[:2]))
ref = model.predict(X[0])
res = model.predict(X[0],
n_tiles=((1, 2, 3) if X[0].ndim == 3 else
(1, 2, 3, 1)))
# assert all(np.allclose(u,v) for u,v in zip(ref,res))
# ask to train only with foreground patches when there are none
# include a constant label image that must trigger a warning
conf.train_foreground_only = 1
conf.train_steps_per_epoch = 1
_X = X[:2]
_Y = [np.zeros_like(Y[0]), np.ones_like(Y[1])]
with pytest.warns(UserWarning):
StarDist3D(conf, name='stardist',
basedir=None).train(_X,
_Y,
validation_data=(X[-1:], Y[-1:]))
def test_model(n_rays, grid):
img = circle_image()
imgs = np.repeat(img[np.newaxis], 10, axis=0)
X = imgs + .6 * np.random.uniform(0, 1, imgs.shape)
Y = imgs.astype(int)
conf = Config2D(n_rays=n_rays,
grid=grid,
use_gpu=False,
train_epochs=1,
train_steps_per_epoch=10,
train_loss_weights=(4, 1),
train_patch_size=(128, 128),
n_channel_in=1)
with tempfile.TemporaryDirectory() as tmp:
model = StarDist2D(conf, name='stardist', basedir=tmp)
model.train(X, Y, validation_data=(X[:3], Y[:3]))
def test_grid(grid,shape):
ss_grid = tuple(slice(0, None, g) for g in grid)
lbl = circle_image(shape=shape, radius=min(shape)//3)
rays = Rays_GoldenSpiral(32)
t1 = time()
d1 = star_dist3D(lbl, rays = rays)
d1 = d1[ss_grid]
t1 = time()-t1
t2 = time()
d2 = star_dist3D(lbl, rays = rays, grid = grid)
t2 = time()-t2
print(f"time 1: {1000*t1:.2f}ms")
print(f"time 2: {1000*t2:.2f}ms")
assert np.allclose(d1,d2)
return lbl, d1, d2
def test_model(tmpdir, n_rays, grid, n_channel, backbone, workers,
use_sequence):
img = circle_image(shape=(64, 80, 96))
imgs = np.repeat(img[np.newaxis], 3, axis=0)
if n_channel is not None:
imgs = np.repeat(imgs[..., np.newaxis], n_channel, axis=-1)
else:
n_channel = 1
X = imgs + .6 * np.random.uniform(0, 1, imgs.shape)
Y = (imgs if imgs.ndim == 4 else imgs[..., 0]).astype(int)
if use_sequence:
X, Y = NumpySequence(X), NumpySequence(Y)
conf = Config3D(
backbone=backbone,
rays=n_rays,
grid=grid,
n_channel_in=n_channel,
use_gpu=False,
train_epochs=1,
train_steps_per_epoch=1,
train_batch_size=2,
train_loss_weights=(4, 1),
train_patch_size=(48, 64, 32),
train_sample_cache=not use_sequence,
)
model = StarDist3D(conf, name='stardist', basedir=str(tmpdir))
model.train(X, Y, validation_data=(X[:2], Y[:2]), workers=workers)
ref = model.predict(X[0])
res = model.predict(X[0],
n_tiles=((1, 2, 3) if X[0].ndim == 3 else
(1, 2, 3, 1)))
# deactivate as order of labels might not be the same
# assert all(np.allclose(u,v) for u,v in zip(ref,res))
return model
mode = "opencl"
rays = Rays_GoldenSpiral(n_rays)
gt = star_dist3D(img, rays=rays, grid=grid, mode=mode)
for dtype in (np.int8, np.int16, np.int32,
np.uint8, np.uint16, np.uint32):
x = star_dist3D(img.astype(dtype), rays=rays, grid=grid, mode=mode)
print("test_stardist3D (mode {mode}) for shape {img.shape} and type {dtype}".format(mode =mode, img = img, dtype = dtype))
check_similar(gt, x)
@pytest.mark.gpu
@pytest.mark.parametrize('img', (real_image3d()[1], random_image((33, 44, 55))))
@pytest.mark.parametrize('n_rays', (4, 16, 32))
@pytest.mark.parametrize('grid', ((1,1,1),(1,2,4)))
def test_cpu_gpu(img, n_rays, grid):
rays = Rays_GoldenSpiral(n_rays)
s_cpp = star_dist3D(img, rays=rays, grid=grid, mode="cpp")
s_ocl = star_dist3D(img, rays=rays, grid=grid, mode="opencl")
check_similar(s_cpp, s_ocl)
if __name__ == '__main__':
from utils import circle_image
rays = Rays_GoldenSpiral(4)
lbl = circle_image((64,) * 3)
a = star_dist3D(lbl, rays=rays, grid=(1, 2, 2), mode="cpp")
b = star_dist3D(lbl, rays=rays, grid=(1, 2, 2), mode="opencl")
print(np.amax(np.abs(a - b)))
