search

mercedes-benz / selfsupervised_flow

Code for Paper "Self-Supervised LiDAR Scene Flow and Motion Segmentation"
MIT License
57 stars 9 forks source link

What is the correct way to train the models (using nuscenes/kitti scene flow dataset)? #7

Open ry4nzhu opened 2 years ago

ry4nzhu commented 2 years ago

Thanks for sharing the code for this cool project. I followed the instructions to setup the codebase, created symlinks for the datasets by ln -s /path/to/nuscenes/dataset ./datasets/nuscenes and installed dependencies.

However, when I tried to run the training (I used the example command python unsup_flow/cli.py --prod -c nuscenes sota_us sota_net no_stataggr_weight), I got the following error:

Traceback (most recent call last):
  File "unsup_flow/cli.py", line 130, in <module>
    sys.exit(main_tf2())  # pragma: no cover
  File "unsup_flow/cli.py", line 57, in main_tf2
    exp.prepare()
  File "./unsup_flow/experiment/experiment.py", line 194, in prepare
    self.cfg.data.name,
  File "./usfl_io/io_tools.py", line 997, in get_unsupervised_flow_dataset
    data_dir=data_dir,
  File "./usfl_io/io_tools.py", line 609, in get_nuscenes_flow_dataset
    filenames, feature_format = get_filenames_and_feature_format(data_dir)
  File "./tfrecutils/__init__.py", line 380, in get_filenames_and_feature_format
    with open(osp.join(dirname, "tfrecords_format.yml"), "r") as fin:
FileNotFoundError: [Errno 2] No such file or directory: './datasets/nuscenes/usfl/in/prepped_datasets/nuscenes/tfrecords_format.yml'

Can you point me where could be wrong and how to run the training? Thanks!

baurst commented 2 years ago

Hi, thank you for the interest in our project. Before you can run the experiment, you have to create the datasets. For this, use the dataset creation scripts in the folder here: unsup_flow/datasets/ where you can find each create.py script for the respective datasets. Run each script once to convert the datasets into our format. We will update the documentation accordingly, we unfortunately forgot to include this step in the Readme.

yanconglin commented 1 year ago

hi @baurst , Can you please share the docker? I am running numerous issues on my machine when following the instructions. It would be great if you can share a link to the processed nuscenes dataset as well. Appreciate it!

YQChen8 commented 11 months ago

Excuse me After I use python unsup_flow/datasets/kitti/create.py It gives me Traceback (most recent call last): File "unsup_flow/datasets/kitti/create.py", line 97, in <module> main() File "unsup_flow/datasets/kitti/create.py", line 89, in main meta=meta, File "/home/zhijun/catkin_ws/src/my-S-LOAM/scripts/ActiveSceneFlow/selfsupervised_flow/tfrecutils/__init__.py", line 118, in write_tfrecord str(fixed_dims), AssertionError: You did not provide valid fixed dims for pcl_t0: [-1, 4]

LYFFF666 commented 10 months ago

I met this error, too. Traceback (most recent call last): File "unsup_flow/datasets/kitti/create.py", line 97, in main() File "unsup_flow/datasets/kitti/create.py", line 89, in main meta=meta, File "/home/zhijun/catkin_ws/src/my-S-LOAM/scripts/ActiveSceneFlow/selfsupervised_flow/tfrecutils/init.py", line 118, in write_tfrecord str(fixed_dims), AssertionError: You did not provide valid fixed dims for pcl_t0: [-1, 4]

baurst commented 10 months ago

Hey, sorry for the delay. In https://github.com/mercedes-benz/selfsupervised_flow/blob/a51cb5dddb1ab409ce18d3d22d650d712fa60ae1/tfrecutils/__init__.py#L115

Am Di., 7. Nov. 2023 um 19:20 Uhr schrieb Yufei Liu < @.***>:

I met this error, too. Traceback (most recent call last): File "unsup_flow/datasets/kitti/create.py", line 97, in main() File "unsup_flow/datasets/kitti/create.py", line 89, in main meta=meta, File "/home/zhijun/catkin_ws/src/my-S-LOAM/scripts/ActiveSceneFlow/selfsupervised_flow/tfrecutils/ init.py", line 118, in write_tfrecord str(fixed_dims), AssertionError: You did not provide valid fixed dims for pcl_t0: [-1, 4]

— Reply to this email directly, view it on GitHub https://github.com/mercedes-benz/selfsupervised_flow/issues/7#issuecomment-1799400570, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG4FUBNFTTLAXD25GAJEDODYDJ3V5AVCNFSM52RNCEG2U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCNZZHE2DAMBVG4YA . You are receiving this because you were mentioned.Message ID: @.***>

LYFFF666 commented 10 months ago

Thanks for the reply, training is ongoing