Updates for PR comments (6a9bfeca) · Commits · aflab / astrogeology / Autocnet

autocnet/graph/cluster_submit.py

+3 −1

Original line number	Diff line number	Diff line
		@@ -213,7 +213,9 @@ def main(): # pragma: no cover
		args = vars(parse_args())
		# set up the logger
		logging.basicConfig(level=os.environ.get("AUTOCNET_LOGLEVEL", "INFO"))
		# Get the message
		# Get the message; 30s timeout should be long enough to handle minor
		# network issues or congestion without holding onto cluster resources
		# for an undue amount of time.
		queue = StrictRedis(host=args['host'], port=args['port'], db=0,
		socket_timeout=30, socket_connect_timeout=30)
		manage_messages(args, queue)

autocnet/matcher/subpixel.py

+26 −4

Original line number	Diff line number	Diff line
		@@ -805,6 +805,7 @@ def subpixel_register_point_smart(point,
		chooser='subpixel_register_point_smart',
		verbose=False,
		func=pattern_match,
		maximum_affine_reprojective_error=1.0,
		ncg=None):

		"""
		@@ -882,7 +883,8 @@ def subpixel_register_point_smart(point,
		size_y=match_kwargs['image_size'][1])
		try:
		baseline_affine = estimate_local_affine(moving_roi,
		reference_roi)
		reference_roi,
		maximum_affine_reprojective_error)
		except Exception as e:
		log.exception(e)
		m = {'id': measure.id,
		@@ -1121,6 +1123,7 @@ def validate_candidate_measure(point,
		node_cache,
		parameters=[],
		func=pattern_match,
		maximum_affine_reprojective_error=1.0,
		**kwargs):
		"""
		Compute the matching distances, matching the reference measure to the measure
		@@ -1183,7 +1186,9 @@ def validate_candidate_measure(point,
		buffer=20)

		try:
		baseline_affine = estimate_local_affine(moving_roi, reference_roi)
		baseline_affine = estimate_local_affine(moving_roi,
		reference_roi,
		maximum_affine_reprojective_error)
		except:
		log.error('Unable to transform image to reference space. Likely too close to the edge of the non-reference image. Setting ignore=True')
		return [np.inf] * len(parameters)
		@@ -1220,6 +1225,7 @@ def smart_register_point(point,
		parameters=[],
		shared_kwargs={},
		valid_reprojection_distance=1.5,
		maximum_affine_reprojective_error=1.0,
		ncg=None):
		"""
		The entry func for the smart subpixel registration code. This is the user
		@@ -1256,6 +1262,12 @@ def smart_register_point(point,
		measures matched from the moving image to the reference image with a
		distance less than this value in pixels are considered valid. Default: 1.1

		maximum_affine_reprojective_error: float
		reprojective errors less than or equal to this value (default 1.0) are
		masked and removed from affine transform estimations. If less than 3 points
		are found, affine estimation will fail. This value can be used to loosen
		the reprojective constraint estimating the affine transform between
		images.
		Returns
		-------
		measures_to_update : list
		@@ -1269,12 +1281,22 @@ def smart_register_point(point,
		if not isinstance(point, Points):
		point = get_point(ncg, session, point)

		measure_results, node_cache = subpixel_register_point_smart(point, session, ncg=ncg, parameters=parameters, **shared_kwargs)
		measure_results, node_cache = subpixel_register_point_smart(point,
		session,
		ncg=ncg,
		parameters=parameters,
		maximum_affine_reprojective_error=maximum_affine_reprojective_error
		**shared_kwargs)
		measures_to_update, measures_to_set_false = decider(measure_results)
		log.info(f'Found {len(measures_to_update)} measures that found subpixel registration consensus.')
		# Validate that the new position has consensus
		for measure in measures_to_update:
		reprojection_distances = validate_candidate_measure(point, measure, node_cache, parameters=parameters, **shared_kwargs)
		reprojection_distances = validate_candidate_measure(point,
		measure,
		node_cache,
		parameters=parameters,
		maximum_affine_reprojective_error=maximum_affine_reprojective_error,
		**shared_kwargs)
		log.info(f'Validation Distance Boolean: {np.array(reprojection_distances) < valid_reprojection_distance}')
		if np.sum(np.array(reprojection_distances) < valid_reprojection_distance) < 2:
		log.info(f"Measure {measure['id']} failed validation. Setting ignore=True for this measure.")

autocnet/spatial/overlap.py

+14 −6

Original line number	Diff line number	Diff line
		from contextlib import nullcontext
		import time
		import logging
		import warnings

		import shapely
		import json
		@@ -121,8 +120,12 @@ def find_interesting_point(nodes, lon, lat, size=71, **kwargs):

		# Extract ORB features in a sub-image around the desired point
		image_roi = roi.Roi(node.geodata, sample, line)
		try:
		roi_array = image_roi.clip(size_x=size, size_y=size) # Units are pixels for the array

		except Exception as e:
		logging.debug(f'Failed to clip ROI with exception: {e}.')
		log.warn(f'Unable to clip ROI for {node}.')
		continue
		# Check if the image is valid and could be used as the reference
		if not is_valid_lroc_image(roi_array):
		log.info('Failed to find interesting features in image due to poor quality image.')
		@@ -223,7 +226,7 @@ def place_points_in_overlap(overlap,
		candidate_points = compgeom.distribute_points_in_geom(geom, ratio_size=ratio_size, distribute_points_kwargs, kwargs)
		logging.debug(f'Found {len(candidate_points)} in overlap {overlap.id}.')
		if not candidate_points.any():
		warnings.warn(f'Failed to distribute points in overlap {overlap.id}')
		log.warn(f'Failed to distribute points in overlap {overlap.id}')
		return []
		log.info(f'Have {len(candidate_points)} potential points to place in overlap {overlap.id}.')

		@@ -344,8 +347,13 @@ def place_points_in_image(image,
		continue

		# Extract ORB features in a sub-image around the desired point
		try:
		image_roi = roi.Roi(node.geodata, sample, line)
		roi_array = image_roi.clip(size_x=size, size_y=size)
		except Exception as e:
		log.debug(f'Failed to clip ROI with exception: {e}.')
		log.warn(f'Unable to clip ROI for {node}.')
		continue
		interesting = extract_most_interesting(roi_array)

		# kps are in the image space with upper left origin and the roi

autocnet/transformation/affine.py

+26 −19

Original line number	Diff line number	Diff line
		@@ -35,7 +35,8 @@ def estimate_affine_from_sensors(reference_image,
		bcenter_x,
		bcenter_y,
		size_x=40,
		size_y=40):
		size_y=40,
		maximum_affine_reprojective_error=1.0):
		"""
		Using the a priori sensor model, project corner and center points from the reference_image into
		the moving_image and use these points to estimate an affine transformation.
		@@ -54,6 +55,12 @@ def estimate_affine_from_sensors(reference_image,
		half-height of the subimage used in the affine transformation
		size_y: int
		half-width of the subimage used in affine transformation
		maximum_affine_reprojective_error: float
		reprojective errors less than or equal to this value (default 1.0) are
		masked and removed from affine transform estimations. If less than 3 points
		are found, affine estimation will fail. This value can be used to loosen
		the reprojective constraint estimating the affine transform between
		images.

		Returns
		-------
		@@ -72,20 +79,9 @@ def estimate_affine_from_sensors(reference_image,
		base_stopx = int(bcenter_x + size_x)
		base_stopy = int(bcenter_y + size_y)

		match_size = reference_image.raster_size

		# for now, require the entire window resides inside both cubes.
		# if base_stopx > match_size[0]:
		# raise Exception(f"Window: {base_stopx} > {match_size[0]}, center: {bcenter_x},{bcenter_y}")
		# if base_startx < 0:
		# raise Exception(f"Window: {base_startx} < 0, center: {bcenter_x},{bcenter_y}")
		# if base_stopy > match_size[1]:
		# raise Exception(f"Window: {base_stopy} > {match_size[1]}, center: {bcenter_x},{bcenter_y} ")
		# if base_starty < 0:
		# raise Exception(f"Window: {base_starty} < 0, center: {bcenter_x},{bcenter_y}")

		x_coords = [base_startx, base_startx, base_stopx, base_stopx, bcenter_x]
		y_coords = [base_starty, base_stopy, base_stopy, base_starty, bcenter_y]

		# Dispatch to the sensor to get the a priori pixel location in the input image
		lons, lats = reference_image.sensormodel.sampline2lonlat(x_coords, y_coords, allowoutside=True)
		xs, ys = moving_image.sensormodel.lonlat2sampline(lons, lats, allowoutside=True)
		@@ -103,13 +99,14 @@ def estimate_affine_from_sensors(reference_image,
		raise ValueError(f'Unable to find enough points to compute an affine transformation. Found {len(dst_gcps)} points, but need at least 3.')

		log.debug(f'Number of GCPs for affine estimation: {len(dst_gcps)}')

		affine = tf.AffineTransform()
		# Estimate the affine twice. The first time to get an initial estimate
		# and the second time to drop points with an estimated reprojection
		# error greater than or equal to 0.1px.
		# error greater than the passed maximum_affine_reprojective_error.
		affine.estimate(np.array(base_gcps), np.array(dst_gcps))
		residuals = affine.residuals(np.array(base_gcps), np.array(dst_gcps))
		mask = residuals <= 1
		mask = residuals <= maximum_affine_reprojective_error
		if len(np.array(base_gcps)[mask]) < 3:
		logging.info(f'Affine residuals: {residuals}.')
		raise ValueError(f'Unable to find enough points to compute an affine transformation. Found {len(np.array(dst_gcps)[mask])} points, but need at least 3.')
		@@ -122,7 +119,7 @@ def estimate_affine_from_sensors(reference_image,
		return affine


		def estimate_local_affine(reference_roi, moving_roi, size_x=60, size_y=60):
		def estimate_local_affine(reference_roi, moving_roi, size_x=60, size_y=60, maximum_affine_reprojective_error=1.0):
		"""
		Applies the affine transfromation calculated in estimate_affine_from_sensors to the moving region of interest (ROI).

		@@ -135,7 +132,16 @@ def estimate_local_affine(reference_roi, moving_roi, size_x=60, size_y=60):
		moving_image : autocnet.io.geodataset.AGeoDataset
		Image that is expected to move around during the matching process,
		points are projected onto this image to compute an affine

		size_x: int
		half-height of the subimage used in the affine transformation
		size_y: int
		half-width of the subimage used in affine transformation
		maximum_affine_reprojective_error: float
		reprojective errors less than or equal to this value (default 1.0) are
		masked and removed from affine transform estimations. If less than 3 points
		are found, affine estimation will fail. This value can be used to loosen
		the reprojective constraint estimating the affine transform between
		images.
		Returns
		-------
		affine
		@@ -146,7 +152,8 @@ def estimate_local_affine(reference_roi, moving_roi, size_x=60, size_y=60):
		reference_roi.x,
		reference_roi.y,
		size_x=size_x,
		size_y=size_y)
		size_y=size_y,
		maximum_affine_reprojective_error=1.0)


		# Remove the translation from the transformation. Users of this function should add