search

TRAILab / CaDDN

Categorical Depth Distribution Network for Monocular 3D Object Detection (CVPR 2021 Oral)
Apache License 2.0
359 stars 62 forks source link

Training on nuscenes dataset #107

Open thom966 opened 2 years ago

thom966 commented 2 years ago

Hello CaDDN author, thank you for your great work.

I am trying to train the model using nuscenes dataset. So, I have modified the config yaml file but I received an error /backbones_3d/ffe/depth_ffe.py", line 51, in forward images = batch_dict["images"] KeyError: 'images'

It seems like the dataloader extracts batch of data for lidar only, not camera front view.

array(['n008-2018-08-28-16-43-51-0400__LIDAR_TOP__1535489298547291.pcd',
       'n015-2018-10-02-10-50-40+0800__LIDAR_TOP__1538448754047572.pcd',
       'n008-2018-08-28-16-43-51-0400__LIDAR_TOP__1535489299547057.pcd',
       'n015-2018-07-24-11-22-45+0800__LIDAR_TOP__1532402941797653.pcd'],
      dtype='<U62')

Does it need explicit code modification in /CaDDN/tools/train_utils/train_utils.py file to train on nuscenes?

Look forward for any suggestions. Thank you for your time.

Ng-Alexjc commented 1 year ago

Hello Thom, may i know how to modify the model config yaml file for nuscenes? i've never done it before. Or where can i get some information about this? thank you.