def readinfo():
global BACKEND
print(BACKEND)
instance = request.json
root_path = Path(instance["root_path"])
response = {"status": "normal"}
BACKEND.root_path = root_path
info_path = Path(instance["info_path"])
dataset_class_name = instance["dataset_class_name"]
BACKEND.dataset = get_dataset_class(dataset_class_name)(
root_path=root_path, info_path=info_path)
BACKEND.image_idxes = list(range(len(BACKEND.dataset)))
response["image_indexes"] = BACKEND.image_idxes
app.logger.info(f'{len(BACKEND.image_idxes)} data')
response = jsonify(results=[response])
response.headers['Access-Control-Allow-Headers'] = '*'
return response
def kitti_gt_fgm_data_prep(old_root_path, old_trainval_info_path,
new_root_path, new_train_info_path):
from second.core import box_np_ops
from second.data.dataset import get_dataset_class
dataset = get_dataset_class('KittiDataset')(
root_path=old_root_path, info_path=old_trainval_info_path)
for i in trange(len(dataset)):
image_idx = i
sensor_data = dataset.get_sensor_data(i)
if 'image_idx' in sensor_data['metadata']:
image_idx = sensor_data['metadata']['image_idx']
points = sensor_data['lidar']['points']
annos = sensor_data['lidar']['annotations']
gt_boxes = annos['boxes']
gt_mask = box_np_ops.points_in_rbbox(points, gt_boxes)
points_aug = np.concatenate(
(points, gt_mask.max(axis=1, keepdims=True)), axis=1)
points_aug = points_aug.astype(np.float32)
velo_file = 'training/velodyne_reduced/%06d.bin' % (image_idx)
with open(f'{new_root_path}/{velo_file}', 'w') as f:
points_aug.tofile(f)
create_groundtruth_database(dataset_class_name='KittiFGMDataset',
data_path=new_root_path,
info_path=new_train_info_path)
def create_groundtruth_database(dataset_class_name,
data_path,
info_path=None,
used_classes=None,
database_save_path=None,
db_info_save_path=None,
relative_path=True,
add_rgb=False,
lidar_only=False,
bev_only=False,
coors_range=None):
dataset = get_dataset_class(dataset_class_name)(
info_path=info_path,
root_path=data_path,
)
root_path = Path(data_path)
if database_save_path is None:
database_save_path = root_path / 'gt_database'
else:
database_save_path = Path(database_save_path)
if db_info_save_path is None:
db_info_save_path = root_path / "kitti_dbinfos_train.pkl"
database_save_path.mkdir(parents=True, exist_ok=True)
all_db_infos = {}
group_counter = 0
for j in prog_bar(list(range(len(dataset)))):
image_idx = j
sensor_data = dataset.get_sensor_data(j)
if "image_idx" in sensor_data["metadata"]:
image_idx = sensor_data["metadata"]["image_idx"]
points = sensor_data["lidar"]["points"]
annos = sensor_data["lidar"]["annotations"]
gt_boxes = annos["boxes"]
names = annos["names"]
group_dict = {}
group_ids = np.full([gt_boxes.shape[0]], -1, dtype=np.int64)
if "group_ids" in annos:
group_ids = annos["group_ids"]
else:
group_ids = np.arange(gt_boxes.shape[0], dtype=np.int64)
difficulty = np.zeros(gt_boxes.shape[0], dtype=np.int32)
if "difficulty" in annos:
difficulty = annos["difficulty"]
num_obj = gt_boxes.shape[0]
point_indices = box_np_ops.points_in_rbbox(points, gt_boxes)
for i in range(num_obj):
filename = f"{image_idx}_{names[i]}_{i}.bin"
filepath = database_save_path / filename
gt_points = points[point_indices[:, i]]
gt_points[:, :3] -= gt_boxes[i, :3]
with open(filepath, 'w') as f:
gt_points.tofile(f)
if (used_classes is None) or names[i] in used_classes:
if relative_path:
db_path = str(database_save_path.stem + "/" + filename)
else:
db_path = str(filepath)
db_info = {
"name": names[i],
"path": db_path,
"image_idx": image_idx,
"gt_idx": i,
"box3d_lidar": gt_boxes[i],
"num_points_in_gt": gt_points.shape[0],
"difficulty": difficulty[i],
# "group_id": -1,
# "bbox": bboxes[i],
}
local_group_id = group_ids[i]
# if local_group_id >= 0:
if local_group_id not in group_dict:
group_dict[local_group_id] = group_counter
group_counter += 1
db_info["group_id"] = group_dict[local_group_id]
if "score" in annos:
db_info["score"] = annos["score"][i]
if names[i] in all_db_infos:
all_db_infos[names[i]].append(db_info)
else:
all_db_infos[names[i]] = [db_info]
for k, v in all_db_infos.items():
print(f"load {len(v)} {k} database infos")
with open(db_info_save_path, 'wb') as f:
pickle.dump(all_db_infos, f)
def build(input_reader_config,
model_config,
training,
voxel_generator,
target_assigner,
multi_gpu=False):
"""Builds a tensor dictionary based on the InputReader config.
Args:
input_reader_config: A input_reader_pb2.InputReader object.
Returns:
A tensor dict based on the input_reader_config.
Raises:
ValueError: On invalid input reader proto.
ValueError: If no input paths are specified.
"""
if not isinstance(input_reader_config, input_reader_pb2.InputReader):
raise ValueError('input_reader_config not of type '
'input_reader_pb2.InputReader.')
prep_cfg = input_reader_config.preprocess
#import pdb; pdb.set_trace()
dataset_cfg = input_reader_config.dataset
num_point_features = model_config.num_point_features
out_size_factor = get_downsample_factor(model_config)
assert out_size_factor > 0
cfg = input_reader_config
db_sampler_cfg = prep_cfg.database_sampler
db_sampler = None
if len(db_sampler_cfg.sample_groups
) > 0 or db_sampler_cfg.database_info_path != "": # enable sample
db_sampler = dbsampler_builder.build(db_sampler_cfg)
#import pdb; pdb.set_trace()
grid_size = voxel_generator.grid_size
feature_map_size = grid_size[:2] // out_size_factor
feature_map_size = [*feature_map_size, 1][::-1]
#print("feature_map_size", feature_map_size)
assert all([n != '' for n in target_assigner.classes
]), "you must specify class_name in anchor_generators."
dataset_cls = get_dataset_class(dataset_cfg.dataset_class_name)
assert dataset_cls.NumPointFeatures >= 3, "you must set this to correct value"
assert dataset_cls.NumPointFeatures == num_point_features, "currently you need keep them same"
prep_func = partial(
prep_pointcloud,
root_path=dataset_cfg.kitti_root_path,
voxel_generator=voxel_generator,
target_assigner=target_assigner,
training=training,
max_voxels=prep_cfg.max_number_of_voxels,
remove_outside_points=False,
remove_unknown=prep_cfg.remove_unknown_examples,
create_targets=training,
shuffle_points=prep_cfg.shuffle_points,
gt_rotation_noise=list(prep_cfg.groundtruth_rotation_uniform_noise),
gt_loc_noise_std=list(prep_cfg.groundtruth_localization_noise_std),
global_rotation_noise=list(prep_cfg.global_rotation_uniform_noise),
global_scaling_noise=list(prep_cfg.global_scaling_uniform_noise),
global_random_rot_range=list(
prep_cfg.global_random_rotation_range_per_object),
global_translate_noise_std=list(prep_cfg.global_translate_noise_std),
db_sampler=db_sampler,
num_point_features=dataset_cls.NumPointFeatures,
anchor_area_threshold=prep_cfg.anchor_area_threshold,
gt_points_drop=prep_cfg.groundtruth_points_drop_percentage,
gt_drop_max_keep=prep_cfg.groundtruth_drop_max_keep_points,
remove_points_after_sample=prep_cfg.remove_points_after_sample,
remove_environment=prep_cfg.remove_environment,
use_group_id=prep_cfg.use_group_id,
out_size_factor=out_size_factor,
multi_gpu=multi_gpu,
min_points_in_gt=prep_cfg.min_num_of_points_in_gt,
random_flip_x=prep_cfg.random_flip_x,
random_flip_y=prep_cfg.random_flip_y,
sample_importance=prep_cfg.sample_importance)
#import pdb; pdb.set_trace()
ret = target_assigner.generate_anchors(feature_map_size)
class_names = target_assigner.classes
anchors_dict = target_assigner.generate_anchors_dict(feature_map_size)
anchors_list = []
for k, v in anchors_dict.items():
anchors_list.append(v["anchors"])
# anchors = ret["anchors"]
anchors = np.concatenate(anchors_list, axis=0)
anchors = anchors.reshape([-1, target_assigner.box_ndim]) # (199888, 7)
assert np.allclose(anchors,
ret["anchors"].reshape(-1, target_assigner.box_ndim))
matched_thresholds = ret["matched_thresholds"]
unmatched_thresholds = ret["unmatched_thresholds"]
anchors_bv = box_np_ops.rbbox2d_to_near_bbox(anchors[:, [0, 1, 3, 4, 6]])
anchor_cache = {
"anchors": anchors,
"anchors_bv": anchors_bv,
"matched_thresholds": matched_thresholds,
"unmatched_thresholds": unmatched_thresholds,
"anchors_dict": anchors_dict,
}
prep_func = partial(prep_func, anchor_cache=anchor_cache)
dataset = dataset_cls(info_path=dataset_cfg.kitti_info_path,
root_path=dataset_cfg.kitti_root_path,
class_names=class_names,
prep_func=prep_func)
#import pdb; pdb.set_trace()
return dataset
def build(input_reader_config,
model_config,
training,
voxel_generator,
target_assigner,
multi_gpu=False):
"""Builds a tensor dictionary based on the InputReader config.
Args:
input_reader_config: A input_reader_pb2.InputReader object.
Returns:
A tensor dict based on the input_reader_config.
Raises:
ValueError: On invalid input reader proto.
ValueError: If no input paths are specified.
"""
if not isinstance(input_reader_config, input_reader_pb2.InputReader):
raise ValueError('input_reader_config not of type '
'input_reader_pb2.InputReader.')
prep_cfg = input_reader_config.preprocess
dataset_cfg = input_reader_config.dataset #kitti_info_path: "/home/lichao/v1.0-mini/infos_train.pkl"
# kitti_root_path: "/home/lichao/v1.0-mini"
# dataset_class_name: "NuScenesDataset"
num_point_features = model_config.num_point_features #4
out_size_factor = get_downsample_factor(model_config) #8
assert out_size_factor > 0
cfg = input_reader_config
db_sampler_cfg = prep_cfg.database_sampler #database_info_path: "/home/lichao/v1.0-mini/kitti_dbinfos_train.pkl"
# sample_groups { name_to_max_num key: "car" value: 30 } } global_random_rotation_range_per_object: 0.0# global_random_rotation_range_per_object: 0.0# rate: 1.0
db_sampler = None
if len(db_sampler_cfg.sample_groups
) > 0 or db_sampler_cfg.database_info_path != "": # enable sample
db_sampler = dbsampler_builder.build(db_sampler_cfg) #加载了gt_base的一些东西
grid_size = voxel_generator.grid_size #[400,400]
feature_map_size = grid_size[:2] // out_size_factor #[50,50]
feature_map_size = [*feature_map_size, 1][::-1] #[50,50]
print("feature_map_size", feature_map_size)
assert all([n != '' for n in target_assigner.classes
]), "you must specify class_name in anchor_generators."
dataset_cls = get_dataset_class(
dataset_cfg.dataset_class_name) # NuScenesDataset
assert dataset_cls.NumPointFeatures >= 3, "you must set this to correct value"
assert dataset_cls.NumPointFeatures == num_point_features, "currently you need keep them same"
prep_func = partial( #pre_func partial 的功能：固定函数参数，返回一个新的函数。
prep_pointcloud, #data\preprocess.py
root_path=dataset_cfg.kitti_root_path,
voxel_generator=voxel_generator, #VoxelGeneratorV2
target_assigner=target_assigner,
training=training,
max_voxels=prep_cfg.max_number_of_voxels, #25000 eval 30000
remove_outside_points=False,
remove_unknown=prep_cfg.remove_unknown_examples,
create_targets=training,
shuffle_points=prep_cfg.shuffle_points,
gt_rotation_noise=list(prep_cfg.groundtruth_rotation_uniform_noise),
gt_loc_noise_std=list(prep_cfg.groundtruth_localization_noise_std),
global_rotation_noise=list(prep_cfg.global_rotation_uniform_noise),
global_scaling_noise=list(prep_cfg.global_scaling_uniform_noise),
global_random_rot_range=list(
prep_cfg.global_random_rotation_range_per_object),
global_translate_noise_std=list(prep_cfg.global_translate_noise_std),
db_sampler=db_sampler,
num_point_features=dataset_cls.NumPointFeatures,
anchor_area_threshold=prep_cfg.anchor_area_threshold,
gt_points_drop=prep_cfg.groundtruth_points_drop_percentage,
gt_drop_max_keep=prep_cfg.groundtruth_drop_max_keep_points,
remove_points_after_sample=prep_cfg.remove_points_after_sample,
remove_environment=prep_cfg.remove_environment,
use_group_id=prep_cfg.use_group_id,
out_size_factor=out_size_factor, #8
multi_gpu=multi_gpu,
min_points_in_gt=prep_cfg.min_num_of_points_in_gt,
random_flip_x=prep_cfg.random_flip_x,
random_flip_y=prep_cfg.random_flip_y,
sample_importance=prep_cfg.sample_importance)
ret = target_assigner.generate_anchors(feature_map_size)
class_names = target_assigner.classes # ['car', 'bicycle', 'bus', 'construction_vehicle', 'motorcycle', 'pedestrian', 'traffic_cone', 'trailer', 'truck', 'barrier']
anchors_dict = target_assigner.generate_anchors_dict(feature_map_size)
anchors_list = []
for k, v in anchors_dict.items():
anchors_list.append(v["anchors"]) #50000个
# anchors = ret["anchors"]
anchors = np.concatenate(anchors_list, axis=0)
anchors = anchors.reshape([-1, target_assigner.box_ndim])
assert np.allclose(anchors,
ret["anchors"].reshape(-1, target_assigner.box_ndim))
matched_thresholds = ret["matched_thresholds"]
unmatched_thresholds = ret["unmatched_thresholds"]
anchors_bv = box_np_ops.rbbox2d_to_near_bbox(
anchors[:, [0, 1, 3, 4, 6]]) #bev_anchor 4维
anchor_cache = {
"anchors": anchors,
"anchors_bv": anchors_bv,
"matched_thresholds": matched_thresholds,
"unmatched_thresholds": unmatched_thresholds,
"anchors_dict": anchors_dict,
}
prep_func = partial(prep_func, anchor_cache=anchor_cache)
dataset = dataset_cls(
info_path=dataset_cfg.kitti_info_path, #数据的路径
root_path=dataset_cfg.kitti_root_path,
class_names=class_names, #10个类
prep_func=prep_func)
return dataset # _nusc_infos <class 'dict'>: {'lidar_path': '/home/lichao/v1.0-mini/samples/LIDAR_TOP/n015-2018-07-24-11-22-45+0800__LIDAR_TOP__1532402927647951.pcd.bin', 'cam_front_path': '/home/lichao/v1.0-mini/samples/CAM_FRONT/n015-2018-07-24-11-22-45+0800__CAM_FRONT__1532402927612460.jpg', 'token': 'ca9a282c9e77460f8360f564131a8af5', 'sweeps': [], 'lidar2ego_translation': [0.943713, 0.0, 1.84023], 'lidar2ego_rotation': [0.7077955119163518, -0.006492242056004365, 0.010646214713995808, -0.7063073142877817], 'ego2global_translation': [411.3039349319818, 1180.8903791765097, 0.0], 'ego2global_rotation': [0.5720320396729045, -0.0016977771610471074, 0.011798001930183783, -0.8201446642457809], 'timestamp': 1532402927647951, 'gt_boxes': array([[ 1.84143850e+01, 5.95160251e+01, 7.69634574e-01,
def create_my_groundtruth_database(dataset_class_name,
data_path,
info_path=None,
used_classes=None,
database_save_path=None,
db_info_save_path=None,
relative_path=True):
dataset = get_dataset_class(dataset_class_name)(
info_path=info_path,
root_path=data_path,
)
root_path = Path(data_path)
if database_save_path is None:
database_save_path = root_path / 'gt_database'
else:
database_save_path = Path(database_save_path)
if db_info_save_path is None:
db_info_save_path = root_path / "my_dbinfos_train.pkl"
database_save_path.mkdir(parents=True, exist_ok=True)
all_db_infos = {}
group_counter = 0
for j in prog_bar(list(range(len(dataset)))):
image_idx = j
sensor_data = dataset.get_sensor_data(j)
if "image_idx" in sensor_data["metadata"]:
image_idx = sensor_data["metadata"]["image_idx"]
points = sensor_data["lidar"]["points"]
annos = sensor_data["lidar"]["annotations"]
gt_boxes = annos["boxes"]
names = annos["names"]
group_dict = {}
group_ids = np.arange(gt_boxes.shape[0], dtype=np.int64)
difficulty = np.zeros(gt_boxes.shape[0], dtype=np.int32)
num_obj = gt_boxes.shape[0]
point_indices = None
if num_obj > 0:
point_indices = box_np_ops.points_in_rbbox(points, gt_boxes)
for i in range(num_obj):
filename = "{}_{}_{}.bin".format(image_idx, names[i], i)
filepath = database_save_path / filename
gt_points = points[point_indices[:, i]]
if gt_points.shape[0] >= 5:
gt_points[:, :3] -= gt_boxes[i, :3]
with open(filepath, 'w') as f:
gt_points.tofile(f)
if (used_classes is None) or names[i] in used_classes:
if relative_path:
db_path = str(database_save_path.stem + "/" + filename)
else:
db_path = str(filepath)
db_info = {
"name": names[i],
"path": db_path,
"image_idx": image_idx,
"gt_idx": i,
"box3d_lidar": gt_boxes[i],
"num_points_in_gt": gt_points.shape[0],
"difficulty": difficulty[i],
}
local_group_id = group_ids[i]
if local_group_id not in group_dict:
group_dict[local_group_id] = group_counter
group_counter += 1
db_info["group_id"] = group_dict[local_group_id]
if names[i] in all_db_infos:
all_db_infos[names[i]].append(db_info)
else:
all_db_infos[names[i]] = [db_info]
for k, v in all_db_infos.items():
print("load {} {} database infos".format(len(v), k))
with open(db_info_save_path, 'wb') as f:
pickle.dump(all_db_infos, f)
def create_groundtruth_database_with_sweep_info(dataset_class_name,
data_path,
info_path=None,
used_classes=None,
database_save_path=None,
db_info_save_path=None,
relative_path=True,
add_rgb=False,
lidar_only=False,
bev_only=False,
coors_range=None):
dataset = get_dataset_class(dataset_class_name)(
info_path=info_path,
root_path=data_path,
)
root_path = Path(data_path)
if database_save_path is None:
database_save_path = root_path / 'gt_database_with_sweep_info'
else:
database_save_path = Path(database_save_path)
if db_info_save_path is None:
db_info_save_path = root_path / "dbinfos_train_with_sweep_info.pkl"
database_save_path.mkdir(parents=True, exist_ok=True)
all_db_infos = {}
group_counter = 0
for j in prog_bar(list(range(len(dataset)))):
image_idx = j
sensor_data = dataset.get_sensor_data(j)
if "image_idx" in sensor_data["metadata"]:
image_idx = sensor_data["metadata"]["image_idx"]
assert("sweep_points_list" in sensor_data["lidar"])
sweep_points_list = sensor_data["lidar"]["sweep_points_list"]
assert("sweep_annotations" in sensor_data["lidar"])
sweep_gt_boxes_list = sensor_data["lidar"]["sweep_annotations"]["boxes_list"]
sweep_gt_tokens_list = sensor_data["lidar"]["sweep_annotations"]["tokens_list"]
sweep_gt_names_list = sensor_data["lidar"]["sweep_annotations"]["names_list"]
# we focus on the objects in the first frame
# and find the bounding box index of the same object in every frame
points = sensor_data["lidar"]["points"]
annos = sensor_data["lidar"]["annotations"]
names = annos["names"]
gt_boxes = annos["boxes"]
attrs = annos["attrs"]
tokens = sweep_gt_tokens_list[0]
# sanity check with redundancy
assert(len(sweep_gt_boxes_list) == len(sweep_gt_tokens_list) == len(sweep_gt_names_list))
assert(len(gt_boxes) == len(attrs) == len(tokens))
assert(np.all(names == sweep_gt_names_list[0]))
assert(np.all(gt_boxes == sweep_gt_boxes_list[0]))
# on every frame, we mask points inside each bounding box
# but we are not looking at every bounding box
sweep_point_indices_list = []
for sweep_points, sweep_gt_boxes in zip(sweep_points_list, sweep_gt_boxes_list):
sweep_point_indices_list.append(box_np_ops.points_in_rbbox(sweep_points, sweep_gt_boxes))
# crop point cloud based on boxes in the current frame
point_indices = box_np_ops.points_in_rbbox(points, gt_boxes)
#
group_dict = {}
group_ids = np.full([gt_boxes.shape[0]], -1, dtype=np.int64)
if "group_ids" in annos:
group_ids = annos["group_ids"]
else:
group_ids = np.arange(gt_boxes.shape[0], dtype=np.int64)
#
difficulty = np.zeros(gt_boxes.shape[0], dtype=np.int32)
if "difficulty" in annos:
difficulty = annos["difficulty"]
num_obj = gt_boxes.shape[0]
for i in range(num_obj):
filename = f"{image_idx}_{names[i]}_{i}.bin"
filepath = database_save_path / filename
# only use non-key frame boxes when the object is moving
if attrs[i] in ['vehicle.moving', 'cycle.with_rider', 'pedestrian.moving']:
gt_points_list = []
for t in range(len(sweep_points_list)):
# fast pass for most frames
if i < len(sweep_gt_tokens_list[t]) and sweep_gt_tokens_list[t][i] == tokens[i]:
box_idx = i
else:
I = np.flatnonzero(tokens[i] == sweep_gt_tokens_list[t])
if len(I) == 0: continue
elif len(I) == 1: box_idx = I[0]
else: raise ValueError('Identical object tokens')
gt_points_list.append(sweep_points_list[t][sweep_point_indices_list[t][:, box_idx]])
gt_points = np.concatenate(gt_points_list, axis=0)[:, [0, 1, 2, 4]]
else:
gt_points = points[point_indices[:, i]]
# offset points based on the bounding box in the current frame
gt_points[:, :3] -= gt_boxes[i, :3]
with open(filepath, 'w') as f:
gt_points.tofile(f)
if (used_classes is None) or names[i] in used_classes:
if relative_path:
db_path = str(database_save_path.stem + "/" + filename)
else:
db_path = str(filepath)
db_info = {
"name": names[i],
"path": db_path,
"image_idx": image_idx,
"gt_idx": i,
"box3d_lidar": gt_boxes[i],
"num_points_in_gt": gt_points.shape[0],
"difficulty": difficulty[i]
}
local_group_id = group_ids[i]
if local_group_id not in group_dict:
group_dict[local_group_id] = group_counter
group_counter += 1
db_info["group_id"] = group_dict[local_group_id]
if "score" in annos:
db_info["score"] = annos["score"][i]
if names[i] in all_db_infos:
all_db_infos[names[i]].append(db_info)
else:
all_db_infos[names[i]] = [db_info]
for k, v in all_db_infos.items():
print(f"load {len(v)} {k} database infos")
with open(db_info_save_path, 'wb') as f:
pickle.dump(all_db_infos, f)
