-
Notifications
You must be signed in to change notification settings - Fork 103
/
lyftdataset.py
1538 lines (1230 loc) · 60.5 KB
/
lyftdataset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# Lyft Dataset SDK.
# Code written by Oscar Beijbom, 2018.
# Licensed under the Creative Commons [see licence.txt]
# Modified by Vladimir Iglovikov 2019.
import json
import math
import os
import sys
import time
from datetime import datetime
from pathlib import Path
from typing import List, Tuple
import cv2
import matplotlib.pyplot as plt
import numpy as np
import sklearn.metrics
from PIL import Image
from matplotlib.axes import Axes
from pyquaternion import Quaternion
from tqdm import tqdm
from lyft_dataset_sdk.utils.data_classes import Box, LidarPointCloud, RadarPointCloud # NOQA
from lyft_dataset_sdk.utils.geometry_utils import BoxVisibility, box_in_image, view_points # NOQA
from lyft_dataset_sdk.utils.map_mask import MapMask
PYTHON_VERSION = sys.version_info[0]
if not PYTHON_VERSION == 3:
raise ValueError("LyftDataset sdk only supports Python version 3.")
class LyftDataset:
"""Database class for Lyft Dataset to help query and retrieve information from the database."""
def __init__(self, data_path: str, json_path: str, verbose: bool = True, map_resolution: float = 0.1):
"""Loads database and creates reverse indexes and shortcuts.
Args:
data_path: Path to the tables and data.
json_path: Path to the folder with json files
verbose: Whether to print status messages during load.
map_resolution: Resolution of maps (meters).
"""
self.data_path = Path(data_path).expanduser().absolute()
self.json_path = Path(json_path)
self.table_names = [
"category",
"attribute",
"visibility",
"instance",
"sensor",
"calibrated_sensor",
"ego_pose",
"log",
"scene",
"sample",
"sample_data",
"sample_annotation",
"map",
]
start_time = time.time()
# Explicitly assign tables to help the IDE determine valid class members.
self.category = self.__load_table__("category", verbose)
self.attribute = self.__load_table__("attribute", verbose)
self.visibility = self.__load_table__("visibility", verbose)
self.instance = self.__load_table__("instance", verbose, missing_ok=True)
self.sensor = self.__load_table__("sensor", verbose)
self.calibrated_sensor = self.__load_table__("calibrated_sensor", verbose)
self.ego_pose = self.__load_table__("ego_pose", verbose)
self.log = self.__load_table__("log", verbose)
self.scene = self.__load_table__("scene", verbose)
self.sample = self.__load_table__("sample", verbose)
self.sample_data = self.__load_table__("sample_data", verbose)
self.sample_annotation = self.__load_table__("sample_annotation", verbose, missing_ok=True)
self.map = self.__load_table__("map", verbose)
# Initialize map mask for each map record.
for map_record in self.map:
if (self.data_path / map_record["filename"]).is_file():
map_record["mask"] = MapMask(self.data_path / map_record["filename"], resolution=map_resolution)
elif (self.data_path / "train_maps" / "map_raster_palo_alto.png").is_file():
map_record["mask"] = MapMask(
self.data_path / "train_maps" / "map_raster_palo_alto.png", resolution=map_resolution
)
else:
raise FileNotFoundError("Could not file map file.")
if verbose:
for table in self.table_names:
print(f"{len(getattr(self, table))} {table},")
print("Done loading in {:.1f} seconds.\n======".format(time.time() - start_time))
# Make reverse indexes for common lookups.
self.__make_reverse_index__(verbose)
# Initialize LyftDatasetExplorer class
self.explorer = LyftDatasetExplorer(self)
def __load_table__(self, table_name: str, verbose: bool = False, missing_ok: bool = False):
"""Loads a table."""
filepath = self.json_path / f"{table_name}.json"
if not filepath.is_file() and missing_ok:
if verbose:
print(f"JSON file {table_name}.json missing, using empty list")
return []
with open(str(filepath)) as f:
table = json.load(f)
return table
def __make_reverse_index__(self, verbose: bool) -> None:
"""De-normalizes database to create reverse indices for common cases.
Args:
verbose: Whether to print outputs.
"""
start_time = time.time()
if verbose:
print("Reverse indexing ...")
# Store the mapping from token to table index for each table.
self._token2ind = dict()
for table in self.table_names:
self._token2ind[table] = dict()
for ind, member in enumerate(getattr(self, table)):
self._token2ind[table][member["token"]] = ind
# Decorate (adds short-cut) sample_annotation table with for category name.
for record in self.sample_annotation:
inst = self.get("instance", record["instance_token"])
record["category_name"] = self.get("category", inst["category_token"])["name"]
# Decorate (adds short-cut) sample_data with sensor information.
for record in self.sample_data:
cs_record = self.get("calibrated_sensor", record["calibrated_sensor_token"])
sensor_record = self.get("sensor", cs_record["sensor_token"])
record["sensor_modality"] = sensor_record["modality"]
record["channel"] = sensor_record["channel"]
# Reverse-index samples with sample_data and annotations.
for record in self.sample:
record["data"] = {}
record["anns"] = []
for record in self.sample_data:
if record["is_key_frame"]:
sample_record = self.get("sample", record["sample_token"])
sample_record["data"][record["channel"]] = record["token"]
for ann_record in self.sample_annotation:
sample_record = self.get("sample", ann_record["sample_token"])
sample_record["anns"].append(ann_record["token"])
# Add reverse indices from log records to map records.
if "log_tokens" not in self.map[0].keys():
raise Exception("Error: log_tokens not in map table. This code is not compatible with the teaser dataset.")
log_to_map = dict()
for map_record in self.map:
for log_token in map_record["log_tokens"]:
log_to_map[log_token] = map_record["token"]
for log_record in self.log:
log_record["map_token"] = log_to_map[log_record["token"]]
if verbose:
print("Done reverse indexing in {:.1f} seconds.\n======".format(time.time() - start_time))
def get(self, table_name: str, token: str) -> dict:
"""Returns a record from table in constant runtime.
Args:
table_name: Table name.
token: Token of the record.
Returns: Table record.
"""
assert table_name in self.table_names, f"Table {table_name} not found"
return getattr(self, table_name)[self.getind(table_name, token)]
def getind(self, table_name: str, token: str) -> int:
"""Returns the index of the record in a table in constant runtime.
Args:
table_name: Table name.
token: The index of the record in table, table is an array.
Returns:
"""
return self._token2ind[table_name][token]
def field2token(self, table_name: str, field: str, query) -> List[str]:
"""Query all records for a certain field value, and returns the tokens for the matching records.
Runs in linear time.
Args:
table_name: Table name.
field: Field name.
query: Query to match against. Needs to type match the content of the query field.
Returns: List of tokens for the matching records.
"""
matches = []
for member in getattr(self, table_name):
if member[field] == query:
matches.append(member["token"])
return matches
def get_sample_data_path(self, sample_data_token: str) -> Path:
"""Returns the path to a sample_data.
Args:
sample_data_token:
Returns:
"""
sd_record = self.get("sample_data", sample_data_token)
return self.data_path / sd_record["filename"]
def get_sample_data(
self,
sample_data_token: str,
box_vis_level: BoxVisibility = BoxVisibility.ANY,
selected_anntokens: List[str] = None,
flat_vehicle_coordinates: bool = False,
) -> Tuple[Path, List[Box], np.array]:
"""Returns the data path as well as all annotations related to that sample_data.
The boxes are transformed into the current sensor's coordinate frame.
Args:
sample_data_token: Sample_data token.
box_vis_level: If sample_data is an image, this sets required visibility for boxes.
selected_anntokens: If provided only return the selected annotation.
flat_vehicle_coordinates: Instead of current sensor's coordinate frame, use vehicle frame which is
aligned to z-plane in world
Returns: (data_path, boxes, camera_intrinsic <np.array: 3, 3>)
"""
# Retrieve sensor & pose records
sd_record = self.get("sample_data", sample_data_token)
cs_record = self.get("calibrated_sensor", sd_record["calibrated_sensor_token"])
sensor_record = self.get("sensor", cs_record["sensor_token"])
pose_record = self.get("ego_pose", sd_record["ego_pose_token"])
data_path = self.get_sample_data_path(sample_data_token)
if sensor_record["modality"] == "camera":
cam_intrinsic = np.array(cs_record["camera_intrinsic"])
image_size = (sd_record["width"], sd_record["height"])
else:
cam_intrinsic = None
image_size = None
# Retrieve all sample annotations and map to sensor coordinate system.
if selected_anntokens is not None:
boxes = list(map(self.get_box, selected_anntokens))
else:
boxes = self.get_boxes(sample_data_token)
# Make list of Box objects including coord system transforms.
box_list = []
for box in boxes:
if flat_vehicle_coordinates:
# Move box to ego vehicle coord system parallel to world z plane
ypr = Quaternion(pose_record["rotation"]).yaw_pitch_roll
yaw = ypr[0]
box.translate(-np.array(pose_record["translation"]))
box.rotate_around_origin(Quaternion(scalar=np.cos(yaw / 2), vector=[0, 0, np.sin(yaw / 2)]).inverse)
else:
# Move box to ego vehicle coord system
box.translate(-np.array(pose_record["translation"]))
box.rotate_around_origin(Quaternion(pose_record["rotation"]).inverse)
# Move box to sensor coord system
box.translate(-np.array(cs_record["translation"]))
box.rotate_around_origin(Quaternion(cs_record["rotation"]).inverse)
if sensor_record["modality"] == "camera" and not box_in_image(
box, cam_intrinsic, image_size, vis_level=box_vis_level
):
continue
box_list.append(box)
return data_path, box_list, cam_intrinsic
def get_box(self, sample_annotation_token: str) -> Box:
"""Instantiates a Box class from a sample annotation record.
Args:
sample_annotation_token: Unique sample_annotation identifier.
Returns:
"""
record = self.get("sample_annotation", sample_annotation_token)
return Box(
record["translation"],
record["size"],
Quaternion(record["rotation"]),
name=record["category_name"],
token=record["token"],
)
def get_boxes(self, sample_data_token: str) -> List[Box]:
"""Instantiates Boxes for all annotation for a particular sample_data record. If the sample_data is a
keyframe, this returns the annotations for that sample. But if the sample_data is an intermediate
sample_data, a linear interpolation is applied to estimate the location of the boxes at the time the
sample_data was captured.
Args:
sample_data_token: Unique sample_data identifier.
Returns:
"""
# Retrieve sensor & pose records
sd_record = self.get("sample_data", sample_data_token)
curr_sample_record = self.get("sample", sd_record["sample_token"])
if curr_sample_record["prev"] == "" or sd_record["is_key_frame"]:
# If no previous annotations available, or if sample_data is keyframe just return the current ones.
boxes = list(map(self.get_box, curr_sample_record["anns"]))
else:
prev_sample_record = self.get("sample", curr_sample_record["prev"])
curr_ann_recs = [self.get("sample_annotation", token) for token in curr_sample_record["anns"]]
prev_ann_recs = [self.get("sample_annotation", token) for token in prev_sample_record["anns"]]
# Maps instance tokens to prev_ann records
prev_inst_map = {entry["instance_token"]: entry for entry in prev_ann_recs}
t0 = prev_sample_record["timestamp"]
t1 = curr_sample_record["timestamp"]
t = sd_record["timestamp"]
# There are rare situations where the timestamps in the DB are off so ensure that t0 < t < t1.
t = max(t0, min(t1, t))
boxes = []
for curr_ann_rec in curr_ann_recs:
if curr_ann_rec["instance_token"] in prev_inst_map:
# If the annotated instance existed in the previous frame, interpolate center & orientation.
prev_ann_rec = prev_inst_map[curr_ann_rec["instance_token"]]
# Interpolate center.
center = [
np.interp(t, [t0, t1], [c0, c1])
for c0, c1 in zip(prev_ann_rec["translation"], curr_ann_rec["translation"])
]
# Interpolate orientation.
rotation = Quaternion.slerp(
q0=Quaternion(prev_ann_rec["rotation"]),
q1=Quaternion(curr_ann_rec["rotation"]),
amount=(t - t0) / (t1 - t0),
)
box = Box(
center,
curr_ann_rec["size"],
rotation,
name=curr_ann_rec["category_name"],
token=curr_ann_rec["token"],
)
else:
# If not, simply grab the current annotation.
box = self.get_box(curr_ann_rec["token"])
boxes.append(box)
return boxes
def box_velocity(self, sample_annotation_token: str, max_time_diff: float = 1.5) -> np.ndarray:
"""Estimate the velocity for an annotation.
If possible, we compute the centered difference between the previous and next frame.
Otherwise we use the difference between the current and previous/next frame.
If the velocity cannot be estimated, values are set to np.nan.
Args:
sample_annotation_token: Unique sample_annotation identifier.
max_time_diff: Max allowed time diff between consecutive samples that are used to estimate velocities.
Returns: <np.float: 3>. Velocity in x/y/z direction in m/s.
"""
current = self.get("sample_annotation", sample_annotation_token)
has_prev = current["prev"] != ""
has_next = current["next"] != ""
# Cannot estimate velocity for a single annotation.
if not has_prev and not has_next:
return np.array([np.nan, np.nan, np.nan])
if has_prev:
first = self.get("sample_annotation", current["prev"])
else:
first = current
if has_next:
last = self.get("sample_annotation", current["next"])
else:
last = current
pos_last = np.array(last["translation"])
pos_first = np.array(first["translation"])
pos_diff = pos_last - pos_first
time_last = 1e-6 * self.get("sample", last["sample_token"])["timestamp"]
time_first = 1e-6 * self.get("sample", first["sample_token"])["timestamp"]
time_diff = time_last - time_first
if has_next and has_prev:
# If doing centered difference, allow for up to double the max_time_diff.
max_time_diff *= 2
if time_diff > max_time_diff:
# If time_diff is too big, don't return an estimate.
return np.array([np.nan, np.nan, np.nan])
else:
return pos_diff / time_diff
def list_categories(self) -> None:
self.explorer.list_categories()
def list_attributes(self) -> None:
self.explorer.list_attributes()
def list_scenes(self) -> None:
self.explorer.list_scenes()
def list_sample(self, sample_token: str) -> None:
self.explorer.list_sample(sample_token)
def render_pointcloud_in_image(
self,
sample_token: str,
dot_size: int = 5,
pointsensor_channel: str = "LIDAR_TOP",
camera_channel: str = "CAM_FRONT",
out_path: str = None,
) -> None:
self.explorer.render_pointcloud_in_image(
sample_token,
dot_size,
pointsensor_channel=pointsensor_channel,
camera_channel=camera_channel,
out_path=out_path,
)
def render_sample(
self,
sample_token: str,
box_vis_level: BoxVisibility = BoxVisibility.ANY,
nsweeps: int = 1,
out_path: str = None,
) -> None:
self.explorer.render_sample(sample_token, box_vis_level, nsweeps=nsweeps, out_path=out_path)
def render_sample_data(
self,
sample_data_token: str,
with_anns: bool = True,
box_vis_level: BoxVisibility = BoxVisibility.ANY,
axes_limit: float = 40,
ax: Axes = None,
nsweeps: int = 1,
out_path: str = None,
underlay_map: bool = False,
):
return self.explorer.render_sample_data(
sample_data_token,
with_anns,
box_vis_level,
axes_limit,
ax,
num_sweeps=nsweeps,
out_path=out_path,
underlay_map=underlay_map,
)
def render_annotation(
self,
sample_annotation_token: str,
margin: float = 10,
view: np.ndarray = np.eye(4),
box_vis_level: BoxVisibility = BoxVisibility.ANY,
out_path: str = None,
) -> None:
self.explorer.render_annotation(sample_annotation_token, margin, view, box_vis_level, out_path)
def render_instance(self, instance_token: str, out_path: str = None) -> None:
self.explorer.render_instance(instance_token, out_path=out_path)
def render_scene(self, scene_token: str, freq: float = 10, imwidth: int = 640, out_path: str = None) -> None:
self.explorer.render_scene(scene_token, freq, image_width=imwidth, out_path=out_path)
def render_scene_channel(
self,
scene_token: str,
channel: str = "CAM_FRONT",
freq: float = 10,
imsize: Tuple[float, float] = (640, 360),
out_path: Path = None,
interactive: bool = True,
verbose: bool = False,
) -> None:
self.explorer.render_scene_channel(
scene_token=scene_token,
channel=channel,
freq=freq,
image_size=imsize,
out_path=out_path,
interactive=interactive,
verbose=verbose,
)
def render_egoposes_on_map(self, log_location: str, scene_tokens: List = None, out_path: str = None) -> None:
self.explorer.render_egoposes_on_map(log_location, scene_tokens, out_path=out_path)
def render_sample_3d_interactive(self, sample_id: str, render_sample: bool = True) -> None:
"""Render 3D visualization of the sample using plotly
Args:
sample_id: Unique sample identifier.
render_sample: call self.render_sample (Render all LIDAR and camera sample_data in
sample along with annotations.)
"""
import pandas as pd
import plotly.graph_objects as go
sample = self.get("sample", sample_id)
sample_data = self.get("sample_data", sample["data"]["LIDAR_TOP"])
pc = LidarPointCloud.from_file(self.data_path / sample_data["filename"])
_, boxes, _ = self.get_sample_data(sample["data"]["LIDAR_TOP"], flat_vehicle_coordinates=False)
if render_sample:
self.render_sample(sample_id)
df_tmp = pd.DataFrame(pc.points[:3, :].T, columns=["x", "y", "z"])
df_tmp["norm"] = np.sqrt(np.power(df_tmp[["x", "y", "z"]].values, 2).sum(axis=1))
scatter = go.Scatter3d(
x=df_tmp["x"],
y=df_tmp["y"],
z=df_tmp["z"],
mode="markers",
marker=dict(size=1, color=df_tmp["norm"], opacity=0.8),
)
x_lines = []
y_lines = []
z_lines = []
def f_lines_add_nones():
x_lines.append(None)
y_lines.append(None)
z_lines.append(None)
ixs_box_0 = [0, 1, 2, 3, 0]
ixs_box_1 = [4, 5, 6, 7, 4]
for box in boxes:
points = view_points(box.corners(), view=np.eye(3), normalize=False)
x_lines.extend(points[0, ixs_box_0])
y_lines.extend(points[1, ixs_box_0])
z_lines.extend(points[2, ixs_box_0])
f_lines_add_nones()
x_lines.extend(points[0, ixs_box_1])
y_lines.extend(points[1, ixs_box_1])
z_lines.extend(points[2, ixs_box_1])
f_lines_add_nones()
for i in range(4):
x_lines.extend(points[0, [ixs_box_0[i], ixs_box_1[i]]])
y_lines.extend(points[1, [ixs_box_0[i], ixs_box_1[i]]])
z_lines.extend(points[2, [ixs_box_0[i], ixs_box_1[i]]])
f_lines_add_nones()
lines = go.Scatter3d(x=x_lines, y=y_lines, z=z_lines, mode="lines", name="lines")
fig = go.Figure(data=[scatter, lines])
fig.update_layout(scene_aspectmode="data")
fig.show()
class LyftDatasetExplorer:
"""Helper class to list and visualize Lyft Dataset data. These are meant to serve as tutorials and templates for
working with the data."""
def __init__(self, lyftd: LyftDataset):
self.lyftd = lyftd
@staticmethod
def get_color(category_name: str) -> Tuple[int, int, int]:
"""Provides the default colors based on the category names.
This method works for the general Lyft Dataset categories, as well as the Lyft Dataset detection categories.
Args:
category_name:
Returns:
"""
if "bicycle" in category_name or "motorcycle" in category_name:
return 255, 61, 99 # Red
elif "vehicle" in category_name or category_name in ["bus", "car", "construction_vehicle", "trailer", "truck"]:
return 255, 158, 0 # Orange
elif "pedestrian" in category_name:
return 0, 0, 230 # Blue
elif "cone" in category_name or "barrier" in category_name:
return 0, 0, 0 # Black
else:
return 255, 0, 255 # Magenta
def list_categories(self) -> None:
"""Print categories, counts and stats."""
print("Category stats")
# Add all annotations
categories = dict()
for record in self.lyftd.sample_annotation:
if record["category_name"] not in categories:
categories[record["category_name"]] = []
categories[record["category_name"]].append(record["size"] + [record["size"][1] / record["size"][0]])
# Print stats
for name, stats in sorted(categories.items()):
stats = np.array(stats)
print(
"{:27} n={:5}, width={:5.2f}\u00B1{:.2f}, len={:5.2f}\u00B1{:.2f}, height={:5.2f}\u00B1{:.2f}, "
"lw_aspect={:5.2f}\u00B1{:.2f}".format(
name[:27],
stats.shape[0],
np.mean(stats[:, 0]),
np.std(stats[:, 0]),
np.mean(stats[:, 1]),
np.std(stats[:, 1]),
np.mean(stats[:, 2]),
np.std(stats[:, 2]),
np.mean(stats[:, 3]),
np.std(stats[:, 3]),
)
)
def list_attributes(self) -> None:
"""Prints attributes and counts."""
attribute_counts = dict()
for record in self.lyftd.sample_annotation:
for attribute_token in record["attribute_tokens"]:
att_name = self.lyftd.get("attribute", attribute_token)["name"]
if att_name not in attribute_counts:
attribute_counts[att_name] = 0
attribute_counts[att_name] += 1
for name, count in sorted(attribute_counts.items()):
print(f"{name}: {count}")
def list_scenes(self) -> None:
"""Lists all scenes with some meta data."""
def ann_count(record):
count = 0
sample = self.lyftd.get("sample", record["first_sample_token"])
while not sample["next"] == "":
count += len(sample["anns"])
sample = self.lyftd.get("sample", sample["next"])
return count
recs = [
(self.lyftd.get("sample", record["first_sample_token"])["timestamp"], record)
for record in self.lyftd.scene
]
for start_time, record in sorted(recs):
start_time = self.lyftd.get("sample", record["first_sample_token"])["timestamp"] / 1000000
length_time = self.lyftd.get("sample", record["last_sample_token"])["timestamp"] / 1000000 - start_time
location = self.lyftd.get("log", record["log_token"])["location"]
desc = record["name"] + ", " + record["description"]
if len(desc) > 55:
desc = desc[:51] + "..."
if len(location) > 18:
location = location[:18]
print(
"{:16} [{}] {:4.0f}s, {}, #anns:{}".format(
desc,
datetime.utcfromtimestamp(start_time).strftime("%y-%m-%d %H:%M:%S"),
length_time,
location,
ann_count(record),
)
)
def list_sample(self, sample_token: str) -> None:
"""Prints sample_data tokens and sample_annotation tokens related to the sample_token."""
sample_record = self.lyftd.get("sample", sample_token)
print(f"Sample: {sample_record['token']}\n")
for sd_token in sample_record["data"].values():
sd_record = self.lyftd.get("sample_data", sd_token)
print(
f"sample_data_token: {sd_token}, mod: {sd_record['sensor_modality']}, channel: {sd_record['channel']}"
)
print("")
for ann_token in sample_record["anns"]:
ann_record = self.lyftd.get("sample_annotation", ann_token)
print(f"sample_annotation_token: {ann_record['token']}, category: {ann_record['category_name']}")
def map_pointcloud_to_image(self, pointsensor_token: str, camera_token: str) -> Tuple:
"""Given a point sensor (lidar/radar) token and camera sample_data token, load point-cloud and map it to
the image plane.
Args:
pointsensor_token: Lidar/radar sample_data token.
camera_token: Camera sample_data token.
Returns: (pointcloud <np.float: 2, n)>, coloring <np.float: n>, image <Image>).
"""
cam = self.lyftd.get("sample_data", camera_token)
pointsensor = self.lyftd.get("sample_data", pointsensor_token)
pcl_path = self.lyftd.data_path / pointsensor["filename"]
if pointsensor["sensor_modality"] == "lidar":
pc = LidarPointCloud.from_file(pcl_path)
else:
pc = RadarPointCloud.from_file(pcl_path)
image = Image.open(str(self.lyftd.data_path / cam["filename"]))
# Points live in the point sensor frame. So they need to be transformed via global to the image plane.
# First step: transform the point-cloud to the ego vehicle frame for the timestamp of the sweep.
cs_record = self.lyftd.get("calibrated_sensor", pointsensor["calibrated_sensor_token"])
pc.rotate(Quaternion(cs_record["rotation"]).rotation_matrix)
pc.translate(np.array(cs_record["translation"]))
# Second step: transform to the global frame.
poserecord = self.lyftd.get("ego_pose", pointsensor["ego_pose_token"])
pc.rotate(Quaternion(poserecord["rotation"]).rotation_matrix)
pc.translate(np.array(poserecord["translation"]))
# Third step: transform into the ego vehicle frame for the timestamp of the image.
poserecord = self.lyftd.get("ego_pose", cam["ego_pose_token"])
pc.translate(-np.array(poserecord["translation"]))
pc.rotate(Quaternion(poserecord["rotation"]).rotation_matrix.T)
# Fourth step: transform into the camera.
cs_record = self.lyftd.get("calibrated_sensor", cam["calibrated_sensor_token"])
pc.translate(-np.array(cs_record["translation"]))
pc.rotate(Quaternion(cs_record["rotation"]).rotation_matrix.T)
# Fifth step: actually take a "picture" of the point cloud.
# Grab the depths (camera frame z axis points away from the camera).
depths = pc.points[2, :]
# Retrieve the color from the depth.
coloring = depths
# Take the actual picture (matrix multiplication with camera-matrix + renormalization).
points = view_points(pc.points[:3, :], np.array(cs_record["camera_intrinsic"]), normalize=True)
# Remove points that are either outside or behind the camera. Leave a margin of 1 pixel for aesthetic reasons.
mask = np.ones(depths.shape[0], dtype=bool)
mask = np.logical_and(mask, depths > 0)
mask = np.logical_and(mask, points[0, :] > 1)
mask = np.logical_and(mask, points[0, :] < image.size[0] - 1)
mask = np.logical_and(mask, points[1, :] > 1)
mask = np.logical_and(mask, points[1, :] < image.size[1] - 1)
points = points[:, mask]
coloring = coloring[mask]
return points, coloring, image
def render_pointcloud_in_image(
self,
sample_token: str,
dot_size: int = 2,
pointsensor_channel: str = "LIDAR_TOP",
camera_channel: str = "CAM_FRONT",
out_path: str = None,
) -> None:
"""Scatter-plots a point-cloud on top of image.
Args:
sample_token: Sample token.
dot_size: Scatter plot dot size.
pointsensor_channel: RADAR or LIDAR channel name, e.g. 'LIDAR_TOP'.
camera_channel: Camera channel name, e.g. 'CAM_FRONT'.
out_path: Optional path to save the rendered figure to disk.
Returns:
"""
sample_record = self.lyftd.get("sample", sample_token)
# Here we just grab the front camera and the point sensor.
pointsensor_token = sample_record["data"][pointsensor_channel]
camera_token = sample_record["data"][camera_channel]
points, coloring, im = self.map_pointcloud_to_image(pointsensor_token, camera_token)
plt.figure(figsize=(9, 16))
plt.imshow(im)
plt.scatter(points[0, :], points[1, :], c=coloring, s=dot_size)
plt.axis("off")
if out_path is not None:
plt.savefig(out_path)
def render_sample(
self, token: str, box_vis_level: BoxVisibility = BoxVisibility.ANY, nsweeps: int = 1, out_path: str = None
) -> None:
"""Render all LIDAR and camera sample_data in sample along with annotations.
Args:
token: Sample token.
box_vis_level: If sample_data is an image, this sets required visibility for boxes.
nsweeps: Number of sweeps for lidar and radar.
out_path: Optional path to save the rendered figure to disk.
Returns:
"""
record = self.lyftd.get("sample", token)
# Separate RADAR from LIDAR and vision.
radar_data = {}
nonradar_data = {}
for channel, token in record["data"].items():
sd_record = self.lyftd.get("sample_data", token)
sensor_modality = sd_record["sensor_modality"]
if sensor_modality in ["lidar", "camera"]:
nonradar_data[channel] = token
else:
radar_data[channel] = token
num_radar_plots = 1 if len(radar_data) > 0 else 0
# Create plots.
n = num_radar_plots + len(nonradar_data)
cols = 2
fig, axes = plt.subplots(int(np.ceil(n / cols)), cols, figsize=(16, 24))
if len(radar_data) > 0:
# Plot radar into a single subplot.
ax = axes[0, 0]
for i, (_, sd_token) in enumerate(radar_data.items()):
self.render_sample_data(
sd_token, with_annotations=i == 0, box_vis_level=box_vis_level, ax=ax, num_sweeps=nsweeps
)
ax.set_title("Fused RADARs")
# Plot camera and lidar in separate subplots.
for (_, sd_token), ax in zip(nonradar_data.items(), axes.flatten()[num_radar_plots:]):
self.render_sample_data(sd_token, box_vis_level=box_vis_level, ax=ax, num_sweeps=nsweeps)
axes.flatten()[-1].axis("off")
plt.tight_layout()
fig.subplots_adjust(wspace=0, hspace=0)
if out_path is not None:
plt.savefig(out_path)
def render_ego_centric_map(self, sample_data_token: str, axes_limit: float = 40, ax: Axes = None) -> None:
"""Render map centered around the associated ego pose.
Args:
sample_data_token: Sample_data token.
axes_limit: Axes limit measured in meters.
ax: Axes onto which to render.
"""
def crop_image(image: np.array, x_px: int, y_px: int, axes_limit_px: int) -> np.array:
x_min = int(x_px - axes_limit_px)
x_max = int(x_px + axes_limit_px)
y_min = int(y_px - axes_limit_px)
y_max = int(y_px + axes_limit_px)
cropped_image = image[y_min:y_max, x_min:x_max]
return cropped_image
sd_record = self.lyftd.get("sample_data", sample_data_token)
# Init axes.
if ax is None:
_, ax = plt.subplots(1, 1, figsize=(9, 9))
sample = self.lyftd.get("sample", sd_record["sample_token"])
scene = self.lyftd.get("scene", sample["scene_token"])
log = self.lyftd.get("log", scene["log_token"])
map = self.lyftd.get("map", log["map_token"])
map_mask = map["mask"]
pose = self.lyftd.get("ego_pose", sd_record["ego_pose_token"])
pixel_coords = map_mask.to_pixel_coords(pose["translation"][0], pose["translation"][1])
scaled_limit_px = int(axes_limit * (1.0 / map_mask.resolution))
mask_raster = map_mask.mask()
cropped = crop_image(mask_raster, pixel_coords[0], pixel_coords[1], int(scaled_limit_px * math.sqrt(2)))
ypr_rad = Quaternion(pose["rotation"]).yaw_pitch_roll
yaw_deg = -math.degrees(ypr_rad[0])
rotated_cropped = np.array(Image.fromarray(cropped).rotate(yaw_deg))
ego_centric_map = crop_image(
rotated_cropped, rotated_cropped.shape[1] / 2, rotated_cropped.shape[0] / 2, scaled_limit_px
)
ax.imshow(
ego_centric_map, extent=[-axes_limit, axes_limit, -axes_limit, axes_limit], cmap="gray", vmin=0, vmax=150
)
def render_sample_data(
self,
sample_data_token: str,
with_annotations: bool = True,
box_vis_level: BoxVisibility = BoxVisibility.ANY,
axes_limit: float = 40,
ax: Axes = None,
num_sweeps: int = 1,
out_path: str = None,
underlay_map: bool = False,
):
"""Render sample data onto axis.
Args:
sample_data_token: Sample_data token.
with_annotations: Whether to draw annotations.
box_vis_level: If sample_data is an image, this sets required visibility for boxes.
axes_limit: Axes limit for lidar and radar (measured in meters).
ax: Axes onto which to render.
num_sweeps: Number of sweeps for lidar and radar.
out_path: Optional path to save the rendered figure to disk.
underlay_map: When set to true, LIDAR data is plotted onto the map. This can be slow.
"""
# Get sensor modality.
sd_record = self.lyftd.get("sample_data", sample_data_token)
sensor_modality = sd_record["sensor_modality"]
if sensor_modality == "lidar":
# Get boxes in lidar frame.
_, boxes, _ = self.lyftd.get_sample_data(
sample_data_token, box_vis_level=box_vis_level, flat_vehicle_coordinates=True
)
# Get aggregated point cloud in lidar frame.
sample_rec = self.lyftd.get("sample", sd_record["sample_token"])
chan = sd_record["channel"]
ref_chan = "LIDAR_TOP"
pc, times = LidarPointCloud.from_file_multisweep(
self.lyftd, sample_rec, chan, ref_chan, num_sweeps=num_sweeps
)
# Compute transformation matrices for lidar point cloud
cs_record = self.lyftd.get("calibrated_sensor", sd_record["calibrated_sensor_token"])
pose_record = self.lyftd.get("ego_pose", sd_record["ego_pose_token"])
vehicle_from_sensor = np.eye(4)
vehicle_from_sensor[:3, :3] = Quaternion(cs_record["rotation"]).rotation_matrix
vehicle_from_sensor[:3, 3] = cs_record["translation"]
ego_yaw = Quaternion(pose_record["rotation"]).yaw_pitch_roll[0]
rot_vehicle_flat_from_vehicle = np.dot(
Quaternion(scalar=np.cos(ego_yaw / 2), vector=[0, 0, np.sin(ego_yaw / 2)]).rotation_matrix,
Quaternion(pose_record["rotation"]).inverse.rotation_matrix,
)
vehicle_flat_from_vehicle = np.eye(4)
vehicle_flat_from_vehicle[:3, :3] = rot_vehicle_flat_from_vehicle
# Init axes.
if ax is None: