charlesq34 / frustum-pointnets

Frustum PointNets for 3D Object Detection from RGB-D Data
Apache License 2.0
1.59k stars 538 forks source link

Loss function for 3D rotation instead of 1D rotation #122

Closed vinayver198 closed 3 years ago

vinayver198 commented 3 years ago

Hi ,

I have a dataset that has 3 rotation angles , I have already done the requisite changes for training but facing some offsets when prediction the 3 angles (heading,pitch and roll) . To solve this problem you have introduced corner loss . I believe i should modify the loss function to fix this. Can @charlesq34 throw some light on this ?