PJLab-ADG / DetZero

[ICCV 2023] DetZero: Rethinking Offboard 3D Object Detection with Long-term Sequential Point Clouds
https://superkoma.github.io/detzero-page/
Apache License 2.0
332 stars 32 forks source link

recall 0.3=(0, 0)/0 recal .7=(0, 0)/0 #41

Open KaiLong1 opened 11 months ago

KaiLong1 commented 11 months ago

Thank you for your wonderful work. While trying, I encountered a very strange problem

When evaluating, all results are 0

Then I checked the data and there was no problem. Self. infos was able to read it normally, but for some reason, the annos of the data became empty.

{'time_stamp': 1510593602140722, 'sample_idx': 20, 'sequencename': '1005081002024129653...0_5333_150', 'pose': array([[ 9.9984854e-...e=float32), 'annos': {'name': array([], dtype='<U10'), 'difficulty': array([], dtype=int64), 'dimensions': array([], shape=(0, ...e=float64), 'location': array([], shape=(0, ...e=float64), 'heading_angles': array([], dtype=float64), 'velocity': array([[ 2.05744581e...934e-02]]), 'obj_ids': array([], dtype='<U22'), 'tracking_difficulty': array([], dtype=int64), 'num_points_in_gt': array([], dtype=int64), ...}, 'num_points_of_each_lidar': [157177, 4063, 3523, 5472, 4389], 'lidar_path': '..s/0020.npy', 'sequence_len': 199}

KaiLong1 commented 11 months ago
微信图片_20231208163928

It seems to be a problem with data parsing, with incorrect data on the left and normal data on the right.

I suspect this may be related to the version I downloaded from the waymo dataset. The version I downloaded is

4fb1af2fad8e67e86b868959f2cb94f

I also discovered a version later. I'm not sure if this version is the waymo data you used. The link is: https://console.cloud.google.com/storage/browser/waymo_open_dataset_v_1_2_0_individual_files/training?pageState=(%22StorageObjectListTable%22:(%22f%22:%22%255B%255D%22))&prefix=&forceOnObjectsSortingFiltering=false

微信图片_20231208164922