def __init__(self, resolution=(320, 240), near=0.01, far=10, client_id=0):
assert client_id >= 0, 'Please provide a client id (0 by default)'
h, w = min(resolution), max(resolution)
self._client_id = client_id
self._near = near
self._far = far
self._shape = (h, w)
self._rgba = np.zeros(self._shape + (4, ), dtype=np.uint8)
self._mask = np.zeros(self._shape, dtype=np.uint8)
self._depth = np.zeros(self._shape, dtype=np.float32)
self._render_options = dict()
self._render_flags = 0
self.mask_link_index(True)
self.casts_shadow(True)
# Transform between standard camera coordinate (z forward) and OPENGL camera coordinate system
wxyz = transforms3d.euler.euler2quat(np.pi / 2, 0, 0, axes='rxyz')
xyzw = [*wxyz[1:], wxyz[0]]
self.TCCGL = Transform(xyzw, (0, 0, 0))
# Set some default parameters
self.set_extrinsic_bullet(target=(0, 0, 0),
distance=1.6,
yaw=90,
pitch=-35,
roll=0)
self.set_intrinsic_fov(90)
def get_state(self):
obs = dict()
# Get images
rgba, mask, depth = self._shot()
rgb = rgba[..., :3]
obs.update(rgb=rgb, mask=mask, depth=depth)
# Get intrinsic, extrinsic parameters with standard conventions.
h, w = self._shape
if self._K is not None:
K = self._K
else:
K = K_from_fov(self._proj_params['fov'])
trans = self._view_params['target']
orn = euler2quat([
self._view_params[k] * np.pi / 180
for k in ('pitch', 'roll', 'yaw')
],
axes='sxyz')
TWCGL = Transform(orn, trans)
TWC = TWCGL * self.TCCGL.inverse()
obs.update(TWC=TWC.toHomogeneousMatrix(),
K=K,
resolution=(self._shape[1], self._shape[0]),
proj_mat=self._proj_mat,
near=self._near,
far=self._far)
return obs
def setup_scene(self, obj_infos):
labels = [obj['name'] for obj in obj_infos]
bodies = self.body_cache.get_bodies_by_labels(labels)
for (obj_info, body) in zip(obj_infos, bodies):
TWO = Transform(obj_info['TWO'])
body.pose = TWO
color = obj_info.get('color', None)
if color is not None:
pb.changeVisualShape(body.body_id, -1, physicsClientId=0, rgbaColor=color)
return bodies
def set_extrinsic_T(self, TWC):
TWC = Transform(TWC)
TWCGL = TWC * self.TCCGL
xyzw = TWCGL.quaternion.coeffs()
wxyz = [xyzw[-1], *xyzw[:-1]]
pitch, roll, yaw = transforms3d.euler.quat2euler(wxyz, axes='sxyz')
yaw = yaw * 180 / np.pi
pitch = pitch * 180 / np.pi
roll = roll * 180 / np.pi
yaw = (yaw % 360 + 360) % 360
distance = 0.0001
self.set_extrinsic_bullet(target=TWCGL.translation, distance=distance,
pitch=pitch, roll=roll, yaw=yaw)
def set_extrinsic_spherical(self, target=(0, 0, 0), rho=0.6, theta=np.pi/4, phi=0, roll=0):
"""
Angles in *radians*.
https://fr.wikipedia.org/wiki/Coordonn%C3%A9es_sph%C3%A9riques#/media/Fichier:Spherical_Coordinates_(Colatitude,_Longitude)_(b).svg
"""
x = rho * np.sin(theta) * np.cos(phi)
y = rho * np.sin(theta) * np.sin(phi)
z = rho * np.cos(theta)
t = np.array([x, y, z])
R = transforms3d.euler.euler2mat(np.pi, theta, phi, axes='sxyz')
R = R @ transforms3d.euler.euler2mat(0, 0, -np.pi/2 + roll, axes='sxyz')
t += np.array(target)
TWC = Transform(R, t)
self.set_extrinsic_T(TWC)
def load_posecnn_results():
results_path = LOCAL_DATA_DIR / 'saved_detections' / 'ycbv_posecnn.pkl'
results = pkl.loads(results_path.read_bytes())
infos, poses, bboxes = [], [], []
l_offsets = (LOCAL_DATA_DIR / 'bop_datasets/ycbv' /
'offsets.txt').read_text().strip().split('\n')
ycb_offsets = dict()
for l_n in l_offsets:
obj_id, offset = l_n[:2], l_n[3:]
obj_id = int(obj_id)
offset = np.array(json.loads(offset)) * 0.001
ycb_offsets[obj_id] = offset
def mat_from_qt(qt):
wxyz = qt[:4].copy().tolist()
xyzw = [*wxyz[1:], wxyz[0]]
t = qt[4:].copy()
return Transform(xyzw, t)
for scene_view_str, result in results.items():
scene_id, view_id = scene_view_str.split('/')
scene_id, view_id = int(scene_id), int(view_id)
n_dets = result['rois'].shape[0]
for n in range(n_dets):
obj_id = result['rois'][:, 1].astype(np.int)[n]
label = f'obj_{obj_id:06d}'
infos.append(
dict(
scene_id=scene_id,
view_id=view_id,
score=result['rois'][n, 1],
label=label,
))
bboxes.append(result['rois'][n, 2:6])
pose = mat_from_qt(result['poses'][n])
offset = ycb_offsets[obj_id]
pose = pose * Transform((0, 0, 0, 1), offset).inverse()
poses.append(pose.toHomogeneousMatrix())
data = tc.PandasTensorCollection(
infos=pd.DataFrame(infos),
poses=torch.as_tensor(np.stack(poses)).float(),
bboxes=torch.as_tensor(np.stack(bboxes)).float(),
).cpu()
return data
def render_images(self, cam_infos, render_depth=False):
cam_obs = []
for cam_info in cam_infos:
K = cam_info['K']
TWC = Transform(cam_info['TWC'])
resolution = cam_info['resolution']
cam = Camera(resolution=resolution, client_id=self.client_id)
cam.set_intrinsic_K(K)
cam.set_extrinsic_T(TWC)
cam_obs_ = cam.get_state()
if self.background_color is not None:
im = cam_obs_['rgb']
mask = cam_obs_['mask']
im[np.logical_or(mask < 0, mask == 255)] = self.background_color
if render_depth:
depth = cam_obs_['depth']
near, far = cam_obs_['near'], cam_obs_['far']
z_n = 2 * depth - 1
z_e = 2 * near * far / (far + near - z_n * (far - near))
z_e[np.logical_or(mask < 0, mask == 255)] = 0.
cam_obs_['depth'] = z_e
cam_obs.append(cam_obs_)
return cam_obs
def mat_from_qt(qt):
wxyz = qt[:4].copy().tolist()
xyzw = [*wxyz[1:], wxyz[0]]
t = qt[4:].copy()
return Transform(xyzw, t)
class Camera:
def __init__(self, resolution=(320, 240), near=0.01, far=10, client_id=0):
assert client_id >= 0, 'Please provide a client id (0 by default)'
h, w = min(resolution), max(resolution)
self._client_id = client_id
self._near = near
self._far = far
self._shape = (h, w)
self._rgba = np.zeros(self._shape + (4, ), dtype=np.uint8)
self._mask = np.zeros(self._shape, dtype=np.uint8)
self._depth = np.zeros(self._shape, dtype=np.float32)
self._render_options = dict()
self._render_flags = 0
self.mask_link_index(True)
self.casts_shadow(True)
# Transform between standard camera coordinate (z forward) and OPENGL camera coordinate system
wxyz = transforms3d.euler.euler2quat(np.pi / 2, 0, 0, axes='rxyz')
xyzw = [*wxyz[1:], wxyz[0]]
self.TCCGL = Transform(xyzw, (0, 0, 0))
# Set some default parameters
self.set_extrinsic_bullet(target=(0, 0, 0),
distance=1.6,
yaw=90,
pitch=-35,
roll=0)
self.set_intrinsic_fov(90)
def set_extrinsic_bullet(self, target, distance, yaw, pitch, roll):
"""
Angles in *degrees*.
"""
up = 'z'
self._view_params = dict(yaw=yaw,
pitch=pitch,
roll=roll,
target=target,
distance=distance)
self._view_mat = pb.computeViewMatrixFromYawPitchRoll(
target, distance, yaw, pitch, roll, 'xyz'.index(up))
def set_extrinsic_T(self, TWC):
TWC = Transform(TWC)
TWCGL = TWC * self.TCCGL
xyzw = TWCGL.quaternion.coeffs()
wxyz = [xyzw[-1], *xyzw[:-1]]
pitch, roll, yaw = transforms3d.euler.quat2euler(wxyz, axes='sxyz')
yaw = yaw * 180 / np.pi
pitch = pitch * 180 / np.pi
roll = roll * 180 / np.pi
yaw = (yaw % 360 + 360) % 360
distance = 0.0001
self.set_extrinsic_bullet(target=TWCGL.translation,
distance=distance,
pitch=pitch,
roll=roll,
yaw=yaw)
def set_extrinsic_spherical(self,
target=(0, 0, 0),
rho=0.6,
theta=np.pi / 4,
phi=0,
roll=0):
"""
Angles in *radians*.
https://fr.wikipedia.org/wiki/Coordonn%C3%A9es_sph%C3%A9riques#/media/Fichier:Spherical_Coordinates_(Colatitude,_Longitude)_(b).svg
"""
x = rho * np.sin(theta) * np.cos(phi)
y = rho * np.sin(theta) * np.sin(phi)
z = rho * np.cos(theta)
t = np.array([x, y, z])
R = transforms3d.euler.euler2mat(np.pi, theta, phi, axes='sxyz')
R = R @ transforms3d.euler.euler2mat(
0, 0, -np.pi / 2 + roll, axes='sxyz')
t += np.array(target)
TWC = Transform(R, t)
self.set_extrinsic_T(TWC)
def set_intrinsic_K(self, K):
h, w = self._shape
proj_mat = proj_from_K(K, near=self._near, far=self._far, h=h,
w=w).flatten()
assert np.allclose(proj_mat[11], -1)
self._proj_mat = proj_mat
self._K = K
self._proj_params = None
def set_intrinsic_fov(self, fov):
h, w = self._shape
self._proj_params = dict(fov=fov)
self._proj_mat = pb.computeProjectionMatrixFOV(fov=fov,
aspect=w / h,
nearVal=self._near,
farVal=self._far)
self._K = None
def set_intrinsic_f(self, *args):
if len(args) == 2:
fx, fy = args
raise NotImplementedError
else:
assert len(args) == 1
fy = args[0]
h, w = self._shape
fov_y = np.arctan(h * 0.5 / fy) * 180 / np.pi
fov = fov_y * 2
self.set_intrinsic_fov(fov)
def get_state(self):
obs = dict()
# Get images
rgba, mask, depth = self._shot()
rgb = rgba[..., :3]
obs.update(rgb=rgb, mask=mask, depth=depth)
# Get intrinsic, extrinsic parameters with standard conventions.
h, w = self._shape
if self._K is not None:
K = self._K
else:
K = K_from_fov(self._proj_params['fov'])
trans = self._view_params['target']
orn = euler2quat([
self._view_params[k] * np.pi / 180
for k in ('pitch', 'roll', 'yaw')
],
axes='sxyz')
TWCGL = Transform(orn, trans)
TWC = TWCGL * self.TCCGL.inverse()
obs.update(TWC=TWC.toHomogeneousMatrix(),
K=K,
resolution=(self._shape[1], self._shape[0]),
proj_mat=self._proj_mat,
near=self._near,
far=self._far)
return obs
def _shot(self):
""" Computes a RGB image, a depth buffer and a segmentation mask buffer
with body unique ids of visible objects for each pixel.
"""
h, w = self._shape
renderer = pb.ER_BULLET_HARDWARE_OPENGL
w, h, rgba, depth, mask = pb.getCameraImage(
width=w,
height=h,
projectionMatrix=self._proj_mat,
viewMatrix=self._view_mat,
renderer=renderer,
flags=self._render_flags,
**self._render_options,
physicsClientId=self._client_id)
rgba = np.asarray(rgba, dtype=np.uint8).reshape((h, w, 4))
depth = np.asarray(depth, dtype=np.float32).reshape((h, w))
mask = np.asarray(mask, dtype=np.uint8).reshape((h, w))
return rgba, mask, depth
def _project(self, fov, near, far):
""" Apply camera projection matrix.
Args:
fov (float): Field of view.
near float): Near plane distance.
far (float): Far plane distance.
"""
self.near = near
self.far = far
h, w = self._shape
self._proj_mat = pb.computeProjectionMatrixFOV(fov=fov,
aspect=w / h,
nearVal=near,
farVal=far)
def mask_link_index(self, flag):
""" If is enabled, the mask combines the object unique id and link index
as follows: value = objectUniqueId + ((linkIndex+1)<<24).
"""
if flag:
self._render_flags |= pb.ER_SEGMENTATION_MASK_OBJECT_AND_LINKINDEX
else:
self._render_flags &= ~pb.ER_SEGMENTATION_MASK_OBJECT_AND_LINKINDEX
def casts_shadow(self, flag):
""" 1 for shadows, 0 for no shadows. """
self._render_options['shadow'] = 1 if flag else 0
def __getitem__(self, frame_id):
row = self.frame_index.iloc[frame_id]
scene_id, view_id = row.scene_id, row.view_id
view_id = int(view_id)
view_id_str = f'{view_id:06d}'
scene_id_str = f'{int(scene_id):06d}'
scene_dir = self.base_dir / scene_id_str
rgb_dir = scene_dir / 'rgb'
if not rgb_dir.exists():
rgb_dir = scene_dir / 'gray'
rgb_path = rgb_dir / f'{view_id_str}.png'
if not rgb_path.exists():
rgb_path = rgb_path.with_suffix('.jpg')
if not rgb_path.exists():
rgb_path = rgb_path.with_suffix('.tif')
rgb = np.array(Image.open(rgb_path))
if rgb.ndim == 2:
rgb = np.repeat(rgb[..., None], 3, axis=-1)
rgb = rgb[..., :3]
h, w = rgb.shape[:2]
rgb = torch.as_tensor(rgb)
cam_annotation = self.annotations[scene_id_str]['scene_camera'][str(
view_id)]
if 'cam_R_w2c' in cam_annotation:
RC0 = np.array(cam_annotation['cam_R_w2c']).reshape(3, 3)
tC0 = np.array(cam_annotation['cam_t_w2c']) * 0.001
TC0 = Transform(RC0, tC0)
else:
TC0 = Transform(np.eye(3), np.zeros(3))
K = np.array(cam_annotation['cam_K']).reshape(3, 3)
T0C = TC0.inverse()
T0C = T0C.toHomogeneousMatrix()
camera = dict(T0C=T0C, K=K, TWC=T0C, resolution=rgb.shape[:2])
T0C = TC0.inverse()
objects = []
mask = np.zeros((h, w), dtype=np.uint8)
if 'scene_gt_info' in self.annotations[scene_id_str]:
annotation = self.annotations[scene_id_str]['scene_gt'][str(
view_id)]
n_objects = len(annotation)
visib = self.annotations[scene_id_str]['scene_gt_info'][str(
view_id)]
for n in range(n_objects):
RCO = np.array(annotation[n]['cam_R_m2c']).reshape(3, 3)
tCO = np.array(annotation[n]['cam_t_m2c']) * 0.001
TCO = Transform(RCO, tCO)
T0O = T0C * TCO
T0O = T0O.toHomogeneousMatrix()
obj_id = annotation[n]['obj_id']
name = f'obj_{int(obj_id):06d}'
bbox_visib = np.array(visib[n]['bbox_visib'])
x, y, w, h = bbox_visib
x1 = x
y1 = y
x2 = x + w
y2 = y + h
obj = dict(label=name,
name=name,
TWO=T0O,
T0O=T0O,
visib_fract=visib[n]['visib_fract'],
id_in_segm=n + 1,
bbox=[x1, y1, x2, y2])
objects.append(obj)
mask_path = scene_dir / 'mask_visib' / f'{view_id_str}_all.png'
if mask_path.exists():
mask = np.array(Image.open(mask_path))
else:
for n in range(n_objects):
mask_n = np.array(
Image.open(scene_dir / 'mask_visib' /
f'{view_id_str}_{n:06d}.png'))
mask[mask_n == 255] = n + 1
mask = torch.as_tensor(mask)
if self.load_depth:
depth_path = scene_dir / 'depth' / f'{view_id_str}.png'
if not depth_path.exists():
depth_path = depth_path.with_suffix('.tif')
depth = np.array(inout.load_depth(depth_path))
camera['depth'] = depth * cam_annotation['depth_scale'] / 1000
obs = dict(
objects=objects,
camera=camera,
frame_info=row.to_dict(),
)
return rgb, mask, obs
def pose(self):
pos, orn = self._client.getBasePositionAndOrientation(self._body_id)
return Transform(orn, pos).toHomogeneousMatrix()
