search

aimotive / mm_training

Multimodal model training on aiMotive Dataset
https://openreview.net/forum?id=LW3bRLlY-SA
MIT License
13 stars 4 forks source link

Multimodal Models Training #9

Open PaulSudarshan opened 4 months ago

PaulSudarshan commented 4 months ago

Hi @TamasMatuszka Did the training dataset used for multimodal model training consist of LEFT and RIGHT view FishEye cameras? Or only Front and Back cameras were used for training?

If in case the FishEye images were not included for training, is there any possible reasons of how the model is able to detect objects in FishEye space with heavy distortion?

TamasMatuszka commented 4 months ago

Hi @PaulSudarshan ,

We did not use the side cameras, only front and back cams during training.

You could use fisheye cameras by creating two pinhole images from one fisheye image. This way, the distortion can be mitigated. However, this will increase the GPU memory requirement of the model.