Apollo: Does Apollo trained a CNNSeg model for Velodyne 16?

[HanFa]

Hi Developers,

I am wondering if the CNN segmentation model has been trained for 16-channel LiDAR. There seem to be three models for VLP-16, VLP-64, VLP-128.

➜ cnnseg git:(master) ✗ pwd /apollo/modules/perception/production/data/perception/lidar/models/cnnseg

➜ cnnseg git:(master) ✗ ls velodyne128 velodyne16 velodyne64

Are the models retrained with 16-channel, 64 and 128 dataset respectively?

If we want to use this CNN model with Velodyne 32, shall we retrain the model with the 32-channel dataset? Are there any instructions on how to train a CNN model?

Thanks so much for your attention!

[weidezhang]

unfortunately, cnn training pipeline is no open source yet. I would suggest you can try using vlp64 model on vlp32 data.