Open yyunyu opened 8 months ago
@yyunyu From the method pespective, it can. But you should spent time making balance on radar and camera since the number of radar points is so small.
Thanks for your reply! @junjie18 I tried radar data and found that the network is heavily affected by sparse radar data/features, so I think that concatenate fusion operation in the case of using radar data would allow the image features to go into higher dimensions that makes features are more irrationally distributed in this space. I see, then I will focus on making balance on radar and camera.
Do you tried using this fusion mechanism on radar data? @junjie18
Looking at the fusion mechanism of CMT, it does not seem to be suitable for radar and camera fusion.