def test_run_model(self):
model = DummyModel()
app = Application(model)
x = np.random.rand(1, 128, 128, 1)
y = app._run_model(x)
self.assertEqual(x.shape, y[0].shape)
def test_batch_predict(self):
def predict1(x, batch_size=4):
y = np.random.rand(*x.shape)
return [y]
def predict2(x, batch_size=4):
y = np.random.rand(*x.shape)
return [y] * 2
num_images = [4, 8, 10]
num_pred_heads = [1, 2]
batch_sizes = [1, 4, 5]
prod = product(num_images, num_pred_heads, batch_sizes)
for num_image, num_pred_head, batch_size in prod:
model = DummyModel(n_out=num_pred_head)
app = Application(model)
x = np.random.rand(num_image, 128, 128, 1)
if num_pred_head == 1:
app.model.predict = Mock(side_effect=predict1)
else:
app.model.predict = Mock(side_effect=predict2)
y = app._batch_predict(x, batch_size=batch_size)
assert app.model.predict.call_count == np.ceil(num_image /
batch_size)
self.assertEqual(x.shape, y[0].shape)
if num_pred_head == 2:
self.assertEqual(x.shape, y[1].shape)
def test_untile_output(self):
model = DummyModel()
kwargs = {'model_image_shape': (128, 128, 1)}
app = Application(model, **kwargs)
# No tiling
x = np.random.rand(1, 128, 128, 1)
tiles, tile_info = app._tile_input(x)
y = app._untile_output(tiles, tile_info)
self.assertEqual(x.shape, y.shape)
# Tiling square
x = np.random.rand(1, 400, 400, 1)
tiles, tile_info = app._tile_input(x)
y = app._untile_output(tiles, tile_info)
self.assertEqual(x.shape, y.shape)
# Tiling rectangle
x = np.random.rand(1, 300, 500, 1)
tiles, tile_info = app._tile_input(x)
y = app._untile_output(tiles, tile_info)
self.assertEqual(x.shape, y.shape)
# Smaller than expected
x = np.random.rand(1, 100, 100, 1)
tiles, tile_info = app._tile_input(x)
y = app._untile_output(tiles, tile_info)
self.assertEqual(x.shape, y.shape)
def test_predict_segmentation(self):
model = DummyModel()
app = Application(model)
x = np.random.rand(1, 128, 128, 1)
y = app._predict_segmentation(x)
self.assertEqual(x.shape, y.shape)
def test_predict_notimplemented(self):
model = DummyModel()
kwargs = {'model_mpp': 0.65, 'model_image_shape': (128, 128, 1)}
app = Application(model, **kwargs)
x = np.random.rand(1, 500, 500, 1)
with self.assertRaises(NotImplementedError):
app.predict(x)
def test_resize_output(self):
model = DummyModel()
kwargs = {'model_image_shape': (128, 128, 1)}
app = Application(model, **kwargs)
x = np.random.rand(1, 128, 128, 1)
# x.shape = original_shape --> no resize
y = app._resize_output(x, x.shape)
self.assertEqual(x.shape, y.shape)
# x.shape != original_shape --> resize
original_shape = (1, 500, 500, 1)
y = app._resize_output(x, original_shape)
self.assertEqual(original_shape, y.shape)
def test_tile_input(self):
model = DummyModel()
kwargs = {'model_mpp': 0.65,
'model_image_shape': (128, 128, 1)}
app = Application(model, **kwargs)
# No tiling
x = np.random.rand(1, 128, 128, 1)
y, tile_info = app._tile_input(x)
self.assertEqual(x.shape, y.shape)
self.assertIsInstance(tile_info, dict)
# Tiling square
x = np.random.rand(1, 400, 400, 1)
y, tile_info = app._tile_input(x)
self.assertEqual(kwargs['model_image_shape'][:-1], y.shape[1:-1])
self.assertIsInstance(tile_info, dict)
# Tiling rectangle
x = np.random.rand(1, 300, 500, 1)
y, tile_info = app._tile_input(x)
self.assertEqual(kwargs['model_image_shape'][:-1], y.shape[1:-1])
self.assertIsInstance(tile_info, dict)
# Smaller than expected
x = np.random.rand(1, 100, 100, 1)
y, tile_info = app._tile_input(x)
self.assertEqual(kwargs['model_image_shape'][:-1], y.shape[1:-1])
self.assertIsInstance(tile_info, dict)
def test_resize_output(self):
model = DummyModel()
kwargs = {'model_image_shape': (128, 128, 1)}
app = Application(model, **kwargs)
x = np.random.rand(1, 128, 128, 1)
# x.shape = original_shape --> no resize
y = app._resize_output(x, x.shape)
self.assertEqual(x.shape, y.shape)
# x.shape != original_shape --> resize
original_shape = (1, 500, 500, 1)
y = app._resize_output(x, original_shape)
self.assertEqual(original_shape, y.shape)
# test multiple outputs are also resized
x_list = [x, x]
# x.shape = original_shape --> no resize
y = app._resize_output(x_list, x.shape)
self.assertIsInstance(y, list)
for y_sub in y:
self.assertEqual(x.shape, y_sub.shape)
# x.shape != original_shape --> resize
original_shape = (1, 500, 500, 1)
y = app._resize_output(x_list, original_shape)
self.assertIsInstance(y, list)
for y_sub in y:
self.assertEqual(original_shape, y_sub.shape)
def test_resize(self):
model = DummyModel()
kwargs = {'model_mpp': 0.65,
'model_image_shape': (128, 128, 1)}
app = Application(model, **kwargs)
x = np.random.rand(1, 500, 500, 1)
# image_mpp = None --> No resize
y, original_shape = app._resize_input(x, image_mpp=None)
self.assertEqual(x.shape, y.shape)
self.assertEqual(x.shape, original_shape)
# image_mpp = model_mpp --> No resize
y, original_shape = app._resize_input(x, image_mpp=kwargs['model_mpp'])
self.assertEqual(x.shape, y.shape)
self.assertEqual(x.shape, original_shape)
# image_mpp > model_mpp --> resize
y, original_shape = app._resize_input(x, image_mpp=2.1 * kwargs['model_mpp'])
self.assertEqual(2.1, np.round(x.shape[1] / y.shape[1], decimals=1))
self.assertEqual(x.shape, original_shape)
# image_mpp < model_mpp --> resize
y, original_shape = app._resize_input(x, image_mpp=0.7 * kwargs['model_mpp'])
self.assertEqual(0.7, np.round(x.shape[1] / y.shape[1], decimals=1))
self.assertEqual(x.shape, original_shape)
def test_predict_segmentation(self):
model = DummyModel()
app = Application(model)
x = np.random.rand(1, 128, 128, 1)
y = app._predict_segmentation(x)
self.assertEqual(x.shape, y.shape)
# test with different MPP
model = DummyModel()
app = Application(model)
x = np.random.rand(1, 128, 128, 1)
y = app._predict_segmentation(x, image_mpp=1.3)
self.assertEqual(x.shape, y.shape)
def test_preprocess(self):
def _preprocess(x):
y = np.ones(x.shape)
return y
model = DummyModel()
x = np.random.rand(1, 30, 30, 1)
# Test no preprocess input
app = Application(model)
y = app._preprocess(x)
self.assertAllEqual(x, y)
# Test ones function
kwargs = {'preprocessing_fn': _preprocess}
app = Application(model, **kwargs)
y = app._preprocess(x)
self.assertAllEqual(np.ones(x.shape), y)
# Test bad input
kwargs = {'preprocessing_fn': 'x'}
with self.assertRaises(ValueError):
app = Application(model, **kwargs)
def test_postprocess(self):
def _postprocess(Lx):
y = np.ones(Lx[0].shape)
return y
model = DummyModel()
x = np.random.rand(1, 30, 30, 1)
# No input
app = Application(model)
y = app._postprocess(x)
self.assertAllEqual(x, y)
# Ones
kwargs = {'postprocessing_fn': _postprocess}
app = Application(model, **kwargs)
y = app._postprocess([x])
self.assertAllEqual(np.ones(x.shape), y)
# Bad input
kwargs = {'postprocessing_fn': 'x'}
with self.assertRaises(ValueError):
app = Application(model, **kwargs)
def test_format_model_output(self):
def _format_model_output(Lx):
return {'inner-distance': Lx}
model = DummyModel()
x = np.random.rand(1, 30, 30, 1)
# No function
app = Application(model)
y = app._format_model_output(x)
self.assertAllEqual(x, y)
# single image
kwargs = {'format_model_output_fn': _format_model_output}
app = Application(model, **kwargs)
y = app._format_model_output(x)
self.assertAllEqual(x, y['inner-distance'])