Open connorlee77 opened 3 years ago
Yes, I was able to train models using the OpenPCDet repository. I converted the KittiDataset class and used this calib.txt.
I currently have a submission under review, so I am not willing to share my code yet, but if it gets accepted, we plan to make the code public.
Hope this helps for now.
Interesting. Just to clarify, you were able to train good models using just the intensity data specifically from lidar_hdl64laststereo_left rather than the 3d point clouds in lidar_hdl64_last?
I used lidar_hdl64_strongest
(and lidar_hdl64_last
).
Hey @MartinHahner can you please share your modified kitti_dataset.py, pv_rcnn.yaml/pointpillars.yaml and dataset config files for seeing through fog? Thanks!
Hey @MartinHahner, could you please share your modified kitti_dataset.py, model/dataset config files for the SeeingThroughFog dataset if possible? Many thanks!
Yes, I was able to train models using the OpenPCDet repository. I converted the KittiDataset class and used this calib.txt.
I currently have a submission under review, so I am not willing to share my code yet, but if it gets accepted, we plan to make the code public.
Hope this helps for now.
@barzanisar Do you have a clean version that you can share?
Here is the clean version: https://github.com/barzanisar/OpenPCDet/tree/STF_cc I only trained with pv_rcnn.yaml Other model cfg files need to have matching class names: ['PassengerCar', 'Pedestrian', 'RidableVehicle']
Here is the clean version: https://github.com/barzanisar/OpenPCDet/tree/STF_cc I only trained with pv_rcnn.yaml Other model cfg files need to have matching class names: ['PassengerCar', 'Pedestrian', 'RidableVehicle']
Many thanks for your replies @barzanisar @MartinHahner. It helps a lot!
Should this programme run on Ubuntu System?
Has anyone successfully trained an object detector (yolov3, ssd, etc...) on just lidar_hdl64_last_stereo_left or lidar_hdl64_strong_stereo_left? I'm getting pretty trash results, which I believe is due to the sparsity of the points. Anyone got tips for training?