def scene_synthetic_triangulation() -> synthetic_scene.SyntheticInputData:
np.random.seed(42)
reference = geo.TopocentricConverter(47.0, 6.0, 0)
data = synthetic_examples.synthetic_circle_scene(reference)
maximum_depth = 40
projection_noise = 1.0
gps_noise = 0.1
imu_noise = 1.0
gcp_noise = (0.0, 0.0)
gcps_count = 10
gcps_shift = [10.0, 0.0, 100.0]
return synthetic_scene.SyntheticInputData(
data.get_reconstruction(),
reference,
maximum_depth,
projection_noise,
gps_noise,
imu_noise,
gcp_noise,
False,
gcps_count,
gcps_shift,
)
def load_reference(self) -> geo.TopocentricConverter:
"""Load reference as a topocentric converter."""
with self.io_handler.open_rt(self._reference_lla_path()) as fin:
lla = io.json_load(fin)
return geo.TopocentricConverter(lla["latitude"], lla["longitude"],
lla["altitude"])
def generate_exifs(reconstruction, gps_noise, speed_ms=10):
"""Generate fake exif metadata from the reconstruction."""
previous_pose = None
previous_time = 0
exifs = {}
reference = geo.TopocentricConverter(0, 0, 0)
for shot_name in sorted(reconstruction.shots.keys()):
shot = reconstruction.shots[shot_name]
exif = {}
exif['width'] = shot.camera.width
exif['height'] = shot.camera.height
exif['focal_ratio'] = shot.camera.focal
exif['camera'] = str(shot.camera.id)
exif['make'] = str(shot.camera.id)
pose = shot.pose.get_origin()
if previous_pose is not None:
previous_time += np.linalg.norm(pose-previous_pose)*speed_ms
previous_pose = pose
exif['capture_time'] = previous_time
perturb_points([pose], [gps_noise, gps_noise, gps_noise])
_, _, _, comp = rc.shot_lla_and_compass(shot, reference)
lat, lon, alt = reference.to_lla(*pose)
exif['gps'] = {}
exif['gps']['latitude'] = lat
exif['gps']['longitude'] = lon
exif['gps']['altitude'] = alt
exif['gps']['dop'] = gps_noise
exif['compass'] = {'angle': comp}
exifs[shot_name] = exif
return exifs
def add_cluster_neighbors(positions, labels, centers, max_distance):
reflla = np.mean(positions, 0)
reference = geo.TopocentricConverter(reflla[0], reflla[1], 0)
topocentrics = []
for position in positions:
x, y, z = reference.to_topocentric(position[0], position[1], 0)
topocentrics.append([x, y])
topocentrics = np.array(topocentrics)
topo_tree = spatial.cKDTree(topocentrics)
clusters = []
for label in np.arange(centers.shape[0]):
cluster_indices = np.where(labels == label)[0]
neighbors = []
for i in cluster_indices:
neighbors.extend(
topo_tree.query_ball_point(topocentrics[i], max_distance))
cluster = list(np.union1d(cluster_indices, neighbors))
clusters.append(cluster)
return clusters
def reconstruction_from_json(obj: Dict[str, Any]) -> types.Reconstruction:
"""
Read a reconstruction from a json object
"""
reconstruction = types.Reconstruction()
# Extract cameras
for key, value in obj["cameras"].items():
camera = camera_from_json(key, value)
reconstruction.add_camera(camera)
# Extract camera biases
if "biases" in obj:
for key, value in obj["biases"].items():
transform = bias_from_json(value)
reconstruction.set_bias(key, transform)
# Extract rig models
if "rig_cameras" in obj:
for key, value in obj["rig_cameras"].items():
reconstruction.add_rig_camera(rig_camera_from_json(key, value))
# Extract rig instances from shots
if "rig_instances" in obj:
for key, value in obj["rig_instances"].items():
rig_instance_from_json(reconstruction, key, value)
# Extract shots
rig_shots = rig_instance_camera_per_shot(obj)
for key, value in obj["shots"].items():
shot_in_reconstruction_from_json(
reconstruction,
key,
value,
rig_camera_id=rig_shots[key][1] if key in rig_shots else None,
rig_instance_id=rig_shots[key][0] if key in rig_shots else None,
is_pano_shot=False,
)
# Extract points
if "points" in obj:
for key, value in obj["points"].items():
point_from_json(reconstruction, key, value)
# Extract pano_shots
if "pano_shots" in obj:
for key, value in obj["pano_shots"].items():
shot_in_reconstruction_from_json(
reconstruction, key, value, is_pano_shot=True
)
# Extract reference topocentric frame
if "reference_lla" in obj:
lla = obj["reference_lla"]
reconstruction.reference = geo.TopocentricConverter(
lla["latitude"], lla["longitude"], lla["altitude"]
)
return reconstruction
def scene_synthetic_cube():
np.random.seed(42)
data = synthetic_examples.synthetic_cube_scene()
reference = geo.TopocentricConverter(47.0, 6.0, 0)
reconstruction = data.get_reconstruction()
input_data = synthetic_scene.SyntheticInputData(reconstruction, reference,
40, 0.0, 0.0, False)
return reconstruction, input_data.tracks_manager
def test_get_gps_point() -> None:
reference = geo.TopocentricConverter(0, 0, 0)
exifs = {}
exifs["gps"] = {
"latitude": 0.0001,
"longitude": 0.0001,
"altitude": 100.0,
}
origin, direction = pairs_selection.get_gps_point(exifs, reference)
assert np.allclose(origin, [[11.131, 11.057, 0.0]], atol=1e-3)
assert np.allclose(direction, [[0, 0, 1]])
def reconstruction_from_json(obj: t.Dict[str, t.Any]):
"""
Read a reconstruction from a json object
"""
reconstruction = types.Reconstruction()
# Extract cameras
for key, value in obj["cameras"].items():
camera = camera_from_json(key, value)
reconstruction.add_camera(camera)
# Extract rig models
if "rig_cameras" in obj:
for key, value in obj["rig_cameras"].items():
reconstruction.add_rig_camera(rig_camera_from_json(key, value))
# Extract shots
for key, value in obj["shots"].items():
shot_in_reconstruction_from_json(reconstruction, key, value)
# Extract rig instances from shots
if "rig_instances" in obj:
for key, value in obj["rig_instances"].items():
rig_instance_from_json(reconstruction, key, value)
# Extract points
if "points" in obj:
for key, value in obj["points"].items():
point_from_json(reconstruction, key, value)
# Extract pano_shots
if "pano_shots" in obj:
for key, value in obj["pano_shots"].items():
is_pano_shot = True
shot_in_reconstruction_from_json(reconstruction, key, value,
is_pano_shot)
# Extract main and unit shots
if "main_shot" in obj:
# pyre-fixme[16]: `types.Reconstruction` has no attribute `main_shot`
reconstruction.main_shot = obj["main_shot"]
if "unit_shot" in obj:
# pyre-fixme[16]: `types.Reconstruction` has no attribute `unit_shot`
reconstruction.unit_shot = obj["unit_shot"]
# Extract reference topocentric frame
if "reference_lla" in obj:
lla = obj["reference_lla"]
reconstruction.reference = geo.TopocentricConverter(
lla["latitude"], lla["longitude"], lla["altitude"])
return reconstruction
def test_read_ground_control_points():
text = """
{
"points": [
{
"id": "1",
"position": {
"latitude": 52.519134104,
"longitude": 13.400740745,
"altitude": 12.0792090446
},
"observations": [
{
"shot_id": "01.jpg",
"projection": [0.7153, 0.5787]
},
{
"shot_id": "02.jpg",
"projection": [0.8085, 0.3831]
}
]
},
{
"id": "2",
"position": {
"latitude": 52.519251158,
"longitude": 13.400502446,
"altitude": 16.7021233002
},
"observations": [
{
"shot_id": "01.jpg",
"projection": [0.2346, 0.4628]
}
]
}
]
}
"""
fp = StringIO(text)
reference = geo.TopocentricConverter(52.51913, 13.4007, 0)
points = io.read_ground_control_points(fp, reference)
assert len(points) == 2
a, b = (len(point.observations) for point in points)
assert min(a, b) == 1
assert max(a, b) == 2
def generate_exifs(reconstruction, gps_noise, speed_ms=10):
"""
Return extracted exif information, as dictionary, usually with fields:
================ ===== ===================================
Field Type Description
================ ===== ===================================
width int Width of image, in pixels
height int Height of image, in pixels
focal_prior float Focal length (real) / sensor width
================ ===== ===================================
:param image: Image name, with extension (i.e. 123.jpg)
"""
previous_pose = None
previous_time = 0
exifs = {}
reference = geo.TopocentricConverter(0, 0, 0)
for shot_name in sorted(reconstruction.shots.keys()):
shot = reconstruction.shots[shot_name]
exif = {}
exif['width'] = shot.camera.width
exif['height'] = shot.camera.height
exif['focal_prior'] = shot.camera.focal_prior
exif['camera'] = str(shot.camera.id)
exif['make'] = str(shot.camera.id)
pose = shot.pose.get_origin()
if previous_pose is not None:
previous_time += np.linalg.norm(pose - previous_pose) * speed_ms
previous_pose = pose
exif['capture_time'] = previous_time
perturb_points([pose], [gps_noise, gps_noise, gps_noise])
shot_copy = copy.deepcopy(shot)
shot_copy.pose.set_origin(pose)
lat, lon, alt, comp = rc.shot_lla_and_compass(shot_copy, reference)
exif['gps'] = {}
exif['gps']['latitude'] = lat
exif['gps']['longitude'] = lon
exif['gps']['altitude'] = alt
exif['gps']['dop'] = gps_noise
exif['compass'] = {'angle': comp}
exifs[shot_name] = exif
return exifs
def reconstruction_from_json(obj):
"""
Read a reconstruction from a json object
"""
reconstruction = types.Reconstruction()
# Extract alignment meta-data
reconstruction.alignment = alignment_from_json(obj.get('alignment', None))
# Extract cameras
for key, value in iteritems(obj['cameras']):
camera = camera_from_json(key, value)
reconstruction.add_camera(camera)
# Extract shots
for key, value in iteritems(obj['shots']):
shot = shot_from_json(key, value, reconstruction.cameras)
reconstruction.add_shot(shot)
# Extract points
if 'points' in obj:
for key, value in iteritems(obj['points']):
point = point_from_json(key, value)
reconstruction.add_point(point)
# Extract pano_shots
if 'pano_shots' in obj:
reconstruction.pano_shots = {}
for key, value in iteritems(obj['pano_shots']):
shot = shot_from_json(key, value, reconstruction.cameras)
reconstruction.pano_shots[shot.id] = shot
# Extract main and unit shots
if 'main_shot' in obj:
reconstruction.main_shot = obj['main_shot']
if 'main_shots' in obj:
reconstruction.main_shots = obj['main_shots']
if 'unit_shot' in obj:
reconstruction.unit_shot = obj['unit_shot']
# Extract reference topocentric frame
if 'reference_lla' in obj:
lla = obj['reference_lla']
reconstruction.reference = geo.TopocentricConverter(
lla['latitude'], lla['longitude'], lla['altitude'])
return reconstruction
def test_read_gcp_list():
text = """WGS84
13.400740745 52.519134104 12.0792090446 2335.0 1416.7 01.jpg
13.400740745 52.519134104 12.0792090446 2639.1 938.0 02.jpg
13.400502446 52.519251158 16.7021233002 766.0 1133.1 01.jpg
"""
fp = StringIO(text)
reference = geo.TopocentricConverter(52.51913, 13.4007, 0)
images = ["01.jpg", "02.jpg"]
exif = {i: {"width": 3000, "height": 2000} for i in images}
points = io.read_gcp_list(fp, reference, exif)
assert len(points) == 2
a, b = (len(point.observations) for point in points)
assert min(a, b) == 1
assert max(a, b) == 2
def scene_synthetic_rig() -> synthetic_scene.SyntheticInputData:
np.random.seed(42)
data = synthetic_examples.synthetic_rig_scene()
maximum_depth = 40
projection_noise = 1.0
gps_noise = 0.1
reference = geo.TopocentricConverter(47.0, 6.0, 0)
return synthetic_scene.SyntheticInputData(
data.get_reconstruction(),
reference,
maximum_depth,
projection_noise,
gps_noise,
False,
)
def reconstruction_from_json(obj):
"""
Read a reconstruction from a json object
"""
reconstruction = types.Reconstruction()
# Extract cameras
for key, value in obj["cameras"].items():
camera = camera_from_json(key, value)
reconstruction.add_camera(camera)
# Extract shots
for key, value in obj["shots"].items():
shot_from_json(reconstruction, key, value)
# Extract points
if "points" in obj:
for key, value in obj["points"].items():
point_from_json(reconstruction, key, value)
# Extract pano_shots
if "pano_shots" in obj:
for key, value in obj["pano_shots"].items():
is_pano_shot = True
shot_from_json(reconstruction, key, value, is_pano_shot)
# Extract main and unit shots
if "main_shot" in obj:
reconstruction.main_shot = obj["main_shot"]
if "unit_shot" in obj:
reconstruction.unit_shot = obj["unit_shot"]
# Extract reference topocentric frame
if "reference_lla" in obj:
lla = obj["reference_lla"]
reconstruction.reference = geo.TopocentricConverter(
lla["latitude"], lla["longitude"], lla["altitude"]
)
return reconstruction
def reconstruction_from_json(obj):
"""
Read a reconstruction from a json object
"""
reconstruction = types.Reconstruction()
# Extract cameras
for key, value in iteritems(obj['cameras']):
camera = camera_from_json(key, value)
reconstruction.add_camera(camera)
# Extract shots
for key, value in iteritems(obj['shots']):
shot_from_json(reconstruction, key, value)
# Extract points
if 'points' in obj:
for key, value in iteritems(obj['points']):
point_from_json(reconstruction, key, value)
# Extract pano_shots
if 'pano_shots' in obj:
for key, value in iteritems(obj['pano_shots']):
is_pano_shot = True
shot_from_json(reconstruction, key, value, is_pano_shot)
# Extract main and unit shots
if 'main_shot' in obj:
reconstruction.main_shot = obj['main_shot']
if 'unit_shot' in obj:
reconstruction.unit_shot = obj['unit_shot']
# Extract reference topocentric frame
if 'reference_lla' in obj:
lla = obj['reference_lla']
reconstruction.reference = geo.TopocentricConverter(
lla['latitude'], lla['longitude'], lla['altitude'])
return reconstruction
def invent_reference_from_gps_and_gcp(
data: DataSetBase,
images: Optional[List[str]] = None) -> geo.TopocentricConverter:
lat, lon, alt = 0.0, 0.0, 0.0
wlat, wlon, walt = 0.0, 0.0, 0.0
if images is None:
images = data.images()
for image in images:
d = data.load_exif(image)
if "gps" in d and "latitude" in d["gps"] and "longitude" in d["gps"]:
w = 1.0 / max(0.01, d["gps"].get("dop", 15))
lat += w * d["gps"]["latitude"]
lon += w * d["gps"]["longitude"]
wlat += w
wlon += w
if "altitude" in d["gps"]:
alt += w * d["gps"]["altitude"]
walt += w
if not wlat and not wlon:
for gcp in data.load_ground_control_points():
lat += gcp.lla["latitude"]
lon += gcp.lla["longitude"]
wlat += 1
wlon += 1
if gcp.has_altitude:
alt += gcp.lla["altitude"]
walt += 1
if wlat:
lat /= wlat
if wlon:
lon /= wlon
if walt:
alt /= walt
return geo.TopocentricConverter(lat, lon, 0) # Set altitude manually.
def pairs_and_poses():
np.random.seed(42)
data = synthetic_examples.synthetic_cube_scene()
reconstruction = data.get_reconstruction()
reference = geo.TopocentricConverter(0, 0, 0)
input_data = synthetic_scene.SyntheticInputData(reconstruction, reference,
40, 0.0, 0.0, False)
features, tracks_manager = input_data.features, input_data.tracks_manager
points_keys = list(reconstruction.points.keys())
pairs, poses = defaultdict(list), defaultdict(list)
for im1, im2 in tracks_manager.get_all_pairs_connectivity():
tuples = tracks_manager.get_all_common_observations(im1, im2)
f1 = [p.point for k, p, _ in tuples if k in points_keys]
f2 = [p.point for k, _, p in tuples if k in points_keys]
pairs[im1, im2].append((f1, f2))
pose1 = reconstruction.shots[im1].pose
pose2 = reconstruction.shots[im2].pose
poses[im1, im2] = pose2.relative_to(pose1)
camera = list(reconstruction.cameras.values())[0]
return pairs, poses, camera, features, tracks_manager, reconstruction
def test_eq_geo() -> None:
assert geo.TopocentricConverter(40, 30,
0) == geo.TopocentricConverter(40, 30, 0)
assert geo.TopocentricConverter(40, 32, 0) != geo.TopocentricConverter(
40, 30, 0)
def load_reference(self):
"""Load reference as a topocentric converter."""
lla = self.load_reference_lla()
return geo.TopocentricConverter(
lla['latitude'], lla['longitude'], lla['altitude'])
def load_reference(file_path):
"""Load reference as a topocentric converter."""
lla = load_reference_lla(file_path)
return geo.TopocentricConverter(
lla['latitude'], lla['longitude'], lla['altitude'])
def test_read_write_ground_control_points():
text = """
{
"points": [
{
"id": "1",
"observations": [
{
"shot_id": "01.jpg",
"projection": [0.7153, 0.5787]
},
{
"shot_id": "02.jpg",
"projection": [0.8085, 0.3831]
}
]
},
{
"id": "2",
"position": {
"latitude": 52.519251158,
"longitude": 13.400502446,
"altitude": 16.7021233002
},
"observations": [
{
"shot_id": "01.jpg",
"projection": [0.2346, 0.4628]
}
]
}
]
}
"""
def check_points(points):
assert len(points) == 2
p1, p2 = points
if p1.id != "1":
p1, p2 = p2, p1
assert p1.coordinates.has_value is False
assert len(p1.observations) == 2
assert np.allclose(p2.lla["latitude"], 52.519251158)
assert np.allclose(p2.lla["longitude"], 13.400502446)
assert np.allclose(p2.coordinates.value[2], 16.7021233002)
assert len(p2.observations) == 1
reference = geo.TopocentricConverter(52.51913, 13.4007, 0)
# Read json
fp = StringIO(text)
points = io.read_ground_control_points(fp, reference)
check_points(points)
# Write json and re-read
fwrite = StringIO()
io.write_ground_control_points(points, fwrite, reference)
freread = StringIO(fwrite.getvalue())
points_reread = io.read_ground_control_points(freread, reference)
check_points(points_reread)
def generate_exifs(
reconstruction: types.Reconstruction,
gps_noise: Union[Dict[str, float], float],
causal_gps_noise: bool = False,
) -> Dict[str, Any]:
"""Generate fake exif metadata from the reconstruction."""
speed_ms = 10.0
previous_pose = None
previous_time = 0
exifs = {}
reference = geo.TopocentricConverter(0, 0, 0)
def _gps_dop(shot):
gps_dop = 15
if isinstance(gps_noise, float):
gps_dop = gps_noise
if isinstance(gps_noise, dict):
gps_dop = gps_noise[shot.camera.id]
return gps_dop
per_sequence = defaultdict(list)
for shot_name in sorted(reconstruction.shots.keys()):
shot = reconstruction.shots[shot_name]
exif = {}
exif["width"] = shot.camera.width
exif["height"] = shot.camera.height
exif["camera"] = str(shot.camera.id)
exif["make"] = str(shot.camera.id)
exif["skey"] = str(shot.camera.id)
per_sequence[exif["skey"]].append(shot_name)
if shot.camera.projection_type in ["perspective", "fisheye"]:
exif["focal_ratio"] = shot.camera.focal
pose = shot.pose.get_origin()
if previous_pose is not None:
previous_time += np.linalg.norm(pose - previous_pose) * speed_ms
previous_pose = pose
exif["capture_time"] = previous_time
exifs[shot_name] = exif
for sequence_images in per_sequence.values():
if causal_gps_noise:
sequence_gps_dop = _gps_dop(
reconstruction.shots[sequence_images[0]])
perturbations_2d = generate_causal_noise(2, sequence_gps_dop,
len(sequence_images), 2.0)
for i, shot_name in enumerate(sequence_images):
shot = reconstruction.shots[shot_name]
exif = exifs[shot_name]
pose = shot.pose.get_origin()
if causal_gps_noise:
gps_perturbation = [perturbations_2d[j][i]
for j in range(2)] + [0]
else:
gps_noise = _gps_dop(shot)
gps_perturbation = [gps_noise, gps_noise, 0]
pose = np.array([pose])
perturb_points(pose, gps_perturbation)
pose = pose[0]
_, _, _, comp = rc.shot_lla_and_compass(shot, reference)
lat, lon, alt = reference.to_lla(*pose)
exif["gps"] = {}
exif["gps"]["latitude"] = lat
exif["gps"]["longitude"] = lon
exif["gps"]["altitude"] = alt
exif["gps"]["dop"] = _gps_dop(shot)
exif["compass"] = {"angle": comp}
return exifs
def load_reference(self) -> geo.TopocentricConverter:
"""Load reference as a topocentric converter."""
lla = self.load_reference_lla()
return geo.TopocentricConverter(lla["latitude"], lla["longitude"],
lla["altitude"])