Training on pedestrian and cyclist

[Nourhan27]

Hello,

First, Thanks for this great work. My question is how to train the model on pedestrian and cyclist, because i want to match the results reported by the paper for the three classes (car, pedestrian, cyclist)

[zye1996]

Hello,

First, Thanks for this great work. My question is how to train the model on pedestrian and cyclist, because i want to match the results reported by the paper for the three classes (car, pedestrian, cyclist)

I am also looking into this and I am pretty sure you can achieve this by modifying the config file to change the class_name and the corresponding evaluation method. And I have done this on 3dssd using mmdetection3d framework if you want to have a look at that implementation

[zye1996]

Hello,

First, Thanks for this great work. My question is how to train the model on pedestrian and cyclist, because i want to match the results reported by the paper for the three classes (car, pedestrian, cyclist)

I implemented the Pytorch version predicting all three classes here if you are still interested

[RolandoAvides]

Hello @zye1996, could give any suggestion on how to configure the script in order to train in 3 classes?

Additionally, do you think it is better to train on cars and train on ped&cyc, or train in all 3 classes?

I'm currently using 3DSSD from Open-PCDet and changed the class_names and num_classes. However, I still can't evaluate pedestrians and cyclists.

Thanks in advance!

[wangy12]

Hello,
First, Thanks for this great work. My question is how to train the model on pedestrian and cyclist, because i want to match the results reported by the paper for the three classes (car, pedestrian, cyclist)

I implemented the Pytorch version predicting all three classes here if you are still interested

Hi @zye1996 , Will you please share how you changed the config and evaluation of 3DSSD to support multi-class? I looked at your repo of mmdetection3d but no luck. Thanks in advance!!

[zye1996]

Hello,
First, Thanks for this great work. My question is how to train the model on pedestrian and cyclist, because i want to match the results reported by the paper for the three classes (car, pedestrian, cyclist)

I implemented the Pytorch version predicting all three classes here if you are still interested

Hi @zye1996 , Will you please share how you changed the config and evaluation of 3DSSD to support multi-class? I looked at your repo of mmdetection3d but no luck. Thanks in advance!!

Please reach me through [email protected] and I will add you to the repo