def test_set_camera_intrisics(self):
# overwrite parameters
self._cam_info.intrinsic = np.array([390, 390, 320, 240])
self._cam_info.intrinsics_conversion_mode = 'fov'
K_test = np.asarray([[2666.67, 0, 960], [0, 2666.67, 540], [0, 0, 1]])
# set camera and get matrix
camera.set_camera_info(bpy.context.scene, self._cam.data,
self._cam_info)
K = np.asarray(
camera.get_calibration_matrix(bpy.context.scene, self._cam.data))
# test
npt.assert_almost_equal(
K_test,
K,
decimal=2,
err_msg=
'Error while setting camera from intrinsics with conv. mode "fov"')
# overwrite parameters
self._cam_info.intrinsics_conversion_mode = 'mm'
K_test = np.asarray([[2666.67, 0, 960], [0, 2666.67, 540], [0, 0, 1]])
# set camera and get amtrix
camera.set_camera_info(bpy.context.scene, self._cam.data,
self._cam_info)
K = np.asarray(
camera.get_calibration_matrix(bpy.context.scene, self._cam.data))
# test
npt.assert_almost_equal(
K_test,
K,
decimal=2,
err_msg=
'Error while setting camera from intrinsics with conv. mode "mm"')
def test_set_camera_hfov(self):
"test setting up camera with field of view"
# overwrite parameters
self._cam_info.hfov = 78.694
K_test = np.asarray([[73.2517014, 0, 960], [0, 73.2517014, 540], [0, 0, 1]])
# set camera and get matrix
camera.set_camera_info(bpy.context.scene, self._cam.data, self._cam_info)
K = np.asarray(camera.get_calibration_matrix(bpy.context.scene, self._cam.data))
# test
npt.assert_almost_equal(K_test, K, err_msg='Error while setting camera from hfov')
def test_set_camera_swfl(self):
"test setting up camera with sensor-width and focal length"
# overwrite parameters
self._cam_info.sensor_width = 1.89882275303
self._cam_info.focal_length = 1.158
K_test = np.asarray([[1170.91, 0, 960], [0, 1170.91, 540], [0, 0, 1]])
# set camera and get matrix
camera.set_camera_info(bpy.context.scene, self._cam.data, self._cam_info)
K = np.asarray(camera.get_calibration_matrix(bpy.context.scene, self._cam.data))
# test
npt.assert_almost_equal(K_test, K, decimal=2, err_msg='Error while setting camera from swfl')
def postprocess(self, dirinfo, base_filename, camera, objs, zeroing,
**kwargs):
"""Postprocessing the scene.
This step will compute all the data that is relevant for
PoseRenderResult. This data will then be saved to json. In addition,
postprocessing will fix the filenames generated by blender.
Args:
dirinfo(DynamicStruct): struct with directory and path info
base_filename(str): file name
camera(bpy.types.Camera): active camera object
objs(list): list of target objects
zeroing(np.array): array for zeroing camera rotation
Kwargs Args:
postprocess_config(Configuration): postprocess specific config.
See abr/scenes/baseconfiguration and scene configs for specific configuration values.
"""
# get postprocess specific configs
postprocess_config = kwargs.get(
'postprocess_config',
abr_scenes.BaseConfiguration().postprocess)
# camera matrix
K_cam = np.asarray(
camera_utils.get_calibration_matrix(bpy.context.scene,
camera.data))
# first we update the view-layer to get the updated values in
# translation and rotation
bpy.context.view_layer.update()
# the compositor postprocessing takes care of fixing file names
# and saving the masks filename into objs
self.compositor.postprocess()
# rectify range map into depth
# Blender depth maps asare indeed ranges. Here we convert ranges into depth values
fpath_range = os.path.join(dirinfo.images.range,
f'{base_filename}.exr')
# filenames (ranges are stored as true exr values, depth as 16 bit png)
if not os.path.exists(dirinfo.images.depth):
os.mkdir(dirinfo.images.depth)
fpath_depth = os.path.join(dirinfo.images.depth,
f'{base_filename}.png')
# convert
camera_utils.project_pinhole_range_to_rectified_depth(
fpath_range,
fpath_depth,
res_x=bpy.context.scene.render.resolution_x,
res_y=bpy.context.scene.render.resolution_y,
calibration_matrix=K_cam,
scale=postprocess_config.depth_scale)
# NOTE: this assumes the camera(s) for which the disparity is computed
# is(are) the correct one(s). That is it has the correct baseline according to
# the rendered scene
if postprocess_config.compute_disparity:
# check whether current camera name contains any of the given
# string for parallel setup
if any([
c for c in postprocess_config.parallel_cameras
if c in camera.name
]):
# use precomputed depth if available, otherwise use range map
dirpath = os.path.join(dirinfo.images.base_path, 'disparity')
if not os.path.exists(dirpath):
os.mkdir(dirpath)
fpath_disparity = os.path.join(dirpath, f'{base_filename}.png')
# compute map
camera_utils.compute_disparity_from_z_info(
fpath_depth,
fpath_disparity,
baseline_mm=postprocess_config.
parallel_cameras_baseline_mm,
calibration_matrix=K_cam,
res_x=bpy.context.scene.render.resolution_x,
res_y=bpy.context.scene.render.resolution_y,
scale=postprocess_config.depth_scale)
# compute bounding boxes and save annotations
results_gl = ResultsCollection()
results_cv = ResultsCollection()
for obj in objs:
render_result_gl, render_result_cv = self.build_render_result(
obj, camera, zeroing, postprocess_config.visibility_from_mask)
if obj['visible']:
results_gl.add_result(render_result_gl)
results_cv.add_result(render_result_cv)
# if there's no visible object, add single instance results to have general scene information annotated
if len(results_gl) == 0:
results_gl.add_result(render_result_gl)
if len(results_cv) == 0:
results_cv.add_result(render_result_cv)
self.save_annotations(dirinfo, base_filename, results_gl, results_cv)
