Closed canteen-man closed 5 months ago
Yes, as MapTR, we do not utilise temporal information in MapQR for this version. But, can_bus signals are fused into bev queries , copied from BEVFormer, although I think can_bus contributes little here.
你好,感谢你的回复 是因为bevformer用hybrid 2d的temporal self attention在实际对比试验里没有太大提升,所以就都没有用hybrid 2d的TSA去训练吗。 MapTR和MapTR-V2论文里的指标也都是没有用上hybrid 2d的TSA。 谢谢
Actually, I remember we get some improvements when using temporal self attention, but we do not add this temporal information to keep consistent with MapTR.
谢谢 感谢你的帮助
你好 MapTR和MapTR-V2的queue_lenght都是1,MapQR的config里也是1,所以MapQR也没有实际应用CAN吗 谢谢