problem on the pub_test.py

[JimmyGea]

Hi,thank you for your great work.
l try to follow your work on Getting Started with CenterPoint on nuScenes,when l running this command: python tools/nusc_tracking/pub_test.py --work_dir work_dirs/nusc_centerpoint_voxelnet_0075voxel_fix_bn_z/output/outputs --checkpoint work_dirs/nusc_centerpoint_voxelnet_0075voxel_fix_bn_z/output/infos_val_10sweeps_withvelo_filter_True.json
The following error occurred during the pub_test.

Traceback (most recent call last):
File "tools/nusc_tracking/pub_test.py", line 194, in
eval_tracking()
File "tools/nusc_tracking/pub_test.py", line 161, in eval_tracking
args.root
File "tools/nusc_tracking/pub_test.py", line 179, in eval
metrics_summary = nus_eval.main()
File "/home/xuejunming/CenterPoint/nuscenes-devkit/python-sdk/nuscenes/eval/tracking/evaluate.py", line 205, in main
metrics, metric_data_list = self.evaluate()
File "/home/xuejunming/CenterPoint/nuscenes-devkit/python-sdk/nuscenes/eval/tracking/evaluate.py", line 135, in evaluate
accumulate_class(class_name)
File "/home/xuejunming/CenterPoint/nuscenes-devkit/python-sdk/nuscenes/eval/tracking/evaluate.py", line 131, in accumulate_class
curr_md = curr_ev.accumulate()
File "/home/xuejunming/CenterPoint/nuscenes-devkit/python-sdk/nuscenes/eval/tracking/algo.py", line 156, in accumulate
assert unachieved_thresholds + duplicate_thresholds + len(thresh_metrics) == self.num_thresholds
AssertionError

Thanks for your kind attention and look forward your prompt reply.

[JimmyGea]

If I may ask one more question, how to create infos_train_10sweeps_withvelo_filter_True.json? I can only create infos_val_10sweeps_withvelo_filter_True.json according to your instructions.

[tianweiy]

for the first question, maybe related to #292 (comment) ?

[tianweiy]

to generate detection on training set, modify val anno path to info_train

CenterPoint/configs/nusc/voxelnet/nusc_centerpoint_voxelnet_0075voxel_fix_bn_z_flip.py

Line 171 in d3a248f

val_anno = "data/nuScenes/infos_val_10sweeps_withvelo_filter_True.pkl"

[tianweiy]

you can also directly use prediction from #249

[JimmyGea]

to generate detection on training set, modify val anno path to info_train

CenterPoint/configs/nusc/voxelnet/nusc_centerpoint_voxelnet_0075voxel_fix_bn_z_flip.py

Line 171 in d3a248f

val_anno = "data/nuScenes/infos_val_10sweeps_withvelo_filter_True.pkl"

Thank you!