def scenario_filter(scene_df, keep_senarios=[]):
nusc_map = NuScenesMap(dataroot=full_path,
map_name=scene_df.iloc[0]['scene_location'])
keep_instance_token_list = []
processed_instance_token_list = []
for i, r in scene_df.iterrows():
if r.instance_token in processed_instance_token_list:
continue
processed_instance_token_list.append(r.instance_token)
instance_road_object_traj = r.past_instance_on_road_objects + [
r.current_instance_on_road_objects
] + r.future_instance_on_road_objects
for road_objects in instance_road_object_traj:
# filter ados that passes through intersections #
if 'intersection' in keep_senarios:
if road_objects['road_segment'] is not "":
ro = nusc_map.get('road_segment',
road_objects['road_segment'])
if ro['is_intersection']:
keep_instance_token_list.append(r.instance_token)
continue
#filtered_df = scene_df.drop(del_idx).reset_index(drop=True)
filtered_df = scene_df[scene_df.instance_token.str.contains(
'|'.join(keep_instance_token_list))]
return filtered_df
def nuscenes_map_to_line_representation(
nusc_map: NuScenesMap,
patch: List[float],
realign: bool = False) -> Tuple[torch.Tensor]:
record = nusc_map.get_records_in_patch(patch, ["drivable_area"])
pt1, pt2 = [], []
for da_token in record["drivable_area"]:
da = nusc_map.get("drivable_area", da_token)
for poly in map(nusc_map.extract_polygon, da["polygon_tokens"]):
p1, p2 = get_edges_of_polygon_in_patch(poly, patch)
if len(p1) > 0 and len(p2) > 0:
p1, p2 = preprocess_map_edges(
torch.as_tensor(p1),
torch.as_tensor(p2),
passes=10,
tol=0.1,
)
pt1.append(p1)
pt2.append(p2)
pt1, pt2 = torch.cat(pt1), torch.cat(pt2)
if realign:
pt1, pt2 = realign_map_edges(pt1, pt2, 0.0)
centers = (pt1 + pt2) / 2
centers1 = centers.unsqueeze(1)
centers2 = centers.unsqueeze(0)
dist = (centers2 - centers1).pow(2).sum(dim=-1).sqrt()
for i in range(centers.size(0)):
dist[i, i] = 1e12
very_close = (dist < 0.01).any(dim=-1)
to_remove = []
for i, c in enumerate(very_close):
if c:
to_remove.append(i)
for i, rem in enumerate(to_remove):
rem = rem - i
pt1 = torch.cat([pt1[:rem], pt1[rem + 1:]])
pt2 = torch.cat([pt2[:rem], pt2[rem + 1:]])
return pt1, pt2
ego_pose[0] + patch_radius,
ego_pose[1] + patch_radius,
)
records_in_patch = nusc_map.get_records_in_patch(
box_coords,
layer_names,
'intersect',
)
near_plane = 1e-8
# # Retrieve and render each record.
taken_already = []
index = -1
for layer_name in layer_names:
for token in records_in_patch[layer_name]:
record = nusc_map.get(layer_name, token)
line = nusc_map.extract_line(record['line_token'])
if line.is_empty: # Skip lines without nodes
continue
xs, ys = line.xy
points = np.array([
[record['pose']['tx']],
[
record['pose']['ty'],
],
[record['pose']['tz']],
], )
# Transform into the ego vehicle frame for the timestamp of the image.
points = points - \
np.array(poserecord['translation']).reshape(
