yurongyou / modest Goto Github PK

I used OpenPCDet's visualization method to visualize seed labels, but it didn't look quite right. The 3D boxes are floating on the ground and cannot properly wrap around obstacles.
The point cloud I used is "*. bin" files in data_preprocessing/nuscenes/NUSCENESS_KITTI-FORMAT/training/velodyne, and the seed label is generated according to readme into generatecluster_mask/intermediate_results/nusc_bbox_pp_score_fw_20_r0.3.

Hi,

Thank you for your great work! What is the training time on Lyft and Nuscenes with 4 NVIDIA 3090 GPUs? Thanks!

Hi, I can't find the file 'data_preprocessing/lyft/meta_data/lyft_2019_train_sample_tracks.pkl' mentioned in 'data_preprocessing/lyft/LYFT_PREPROCESSING.md'.
Can I generate the file using the existing code? Or I should generate it using poses myself?
Thanks!

Hi,

thank you for your great work, and thank you for publishing the code! I have a question regarding your baseline MODEST-PP (R0), which you mention in your paper (the one that does not require multiple traversals).

In the paper you write: "The seed labels are constructed by the exact same process as described in section 3, except we replace the edge weights in Equation 4 by spatial proximity: [...] and do not perform any PP-score-based filtering on the clusters generated by DBSCAN"

Does that mean, that PP score is not used at all in this case? Because if we only have a single traversal, I believe we will divide by zero in formula (3) (log(1) = 0).

This leads me to believe, that it is not used at all (and the fact that it is not used here, when affinity_type == '3d_l2_distance' )

Does this mean that for MODEST-PP, you are basically doing DBSCAN for each point cloud individually, but the edge weights are just the nearest neighbor distances?

Thank you very much for your help!

Hello，thank you for this outstanding work！
So how can I run MODEST code on av2 datasets?

Did anyone use this repo successfully on custom dataset ?

Hi, I am using the data_processing/nuscenes/nusc2kitti_boston.py script to convert nuScenes data into KITTI format, but got errors with the information as below:

Traceback (most recent call last):
File "nusc2kitti_boston.py", line 585, in
kc.nuscenes_gt_to_kitti()
File "nusc2kitti_boston.py", line 226, in nuscenes_gt_to_kitti
for idx, (lidar_token, cam_front_token) in tqdm(list(enumerate(sample_tokens))))
File "/usr/local/lib/python3.7/dist-packages/joblib/parallel.py", line 1056, in call
self.retrieve()
File "/usr/local/lib/python3.7/dist-packages/joblib/parallel.py", line 935, in retrieve
self._output.extend(job.get(timeout=self.timeout))
File "/usr/lib/python3.7/multiprocessing/pool.py", line 657, in get
raise self._value
File "/usr/lib/python3.7/multiprocessing/pool.py", line 121, in worker
result = (True, func(*args, **kwds))
File "/usr/local/lib/python3.7/dist-packages/joblib/_parallel_backends.py", line 595, in call
return self.func(*args, **kwargs)
File "/usr/local/lib/python3.7/dist-packages/joblib/parallel.py", line 263, in call
for func, args, kwargs in self.items]
File "/usr/local/lib/python3.7/dist-packages/joblib/parallel.py", line 263, in
for func, args, kwargs in self.items]
File "nusc2kitti_boston.py", line 343, in process_token_to_kitti
for sample_annotation_token in sample_annotation_tokens:
NameError: name 'sample_annotation_tokens' is not defined

I have checked this script and it seems because of a lack of definition of the 'sample_annotation_tokens'.
Could you please have a look at this?
Many thanks.

Hello, I have completed the production of the seed label according to the ReadMe tutorial. How can I visualize these results like the images in the paper?

Hi, I download the pretrained pointrcnn model and tested it on KITTI pointcloud data.
But It didn't predict any 3D labels on the data I tested.

The command I used to test is as follows ( the current directory is OpenPCDet/tools):
python demo.py --cfg_file cfgs/lyft_models/pointrcnn_dynamic_obj.yaml --ckpt prcnn_round_40.pth --data_path <KITTI_DATA_FOLDER>/training_data/velodyne/003334.bin

I also tested other models like SECOND or PointPillars, and both are able to predict some 3D labels on the same data. So I think maybe there is some problem with PointRCNN pretrained model?
Thanks!