search

junjie18 / CMT

[ICCV 2023] Cross Modal Transformer: Towards Fast and Robust 3D Object Detection
Other
308 stars 34 forks source link

有谁复现了CMT的结果吗 #76

Open rubbish001 opened 9 months ago

rubbish001 commented 9 months ago

memory, pos_embed = torch.cat([bev_memory, rv_memory], dim=0), torch.cat([bev_pos_embed, rv_pos_embed], dim=0) 这步代码存在问题，，2个的模态的数据分布不一样，进行cat操作，导致不收敛，我感觉这个方法存在问题