search

tsinghua-rll / VoxelNet-tensorflow

A 3D object detection system for autonomous driving.
MIT License
453 stars 123 forks source link

Augmentation #46

Open ansabsheikh9 opened 6 years ago

ansabsheikh9 commented 6 years ago

Hi @jeasinema, I have noticed after augmantation of type 1, and adding delta Z makes some of the boxes are below ground plane. And some are above the ground plane. is it ok to give training data like this? could you please comment on this

untitled

abhigoku10 commented 5 years ago

@jeasinema hi how did u augment the data can you pls share the steps . i too want ot augment the data but not able to do it

ansabsheikh9 commented 5 years ago

Hi @abhigoku10 here is the function which is used for data augmentation. https://github.com/jeasinema/VoxelNet-tensorflow/blob/d24c3eefc5762e4785ca3be6ca7d1fd0ef16bc61/utils/data_aug.py#L23

abhigoku10 commented 5 years ago

@ansabsheikh9 thanks for the pointers , have u done occlusion augmentation.Can you pls share some details on that

ansabsheikh9 commented 5 years ago

@abhigoku10 I do not have any idea about occolusion augmentation, But here are three types of augmentations. For each iteration, one type of augmentation is performed which is selected based on some probability. First is random orientation of the whole scene, second is random magnification, third is augmenting individual object by rotation and translation in three axis.

abhigoku10 commented 5 years ago

@ansabsheikh9 thanks for the information

abhigoku10 commented 5 years ago

@ansabsheikh9 where you able to draw the bounding on the image??