Sense-GVT / Fast-BEV

Fast-BEV: A Fast and Strong Bird’s-Eye View Perception Baseline
Other
608 stars 92 forks source link

重新训练网络,AP始终为0,无法复现论文精度 #60

Closed Mandylove1993 closed 1 year ago

Mandylove1993 commented 1 year ago

你好,我在云平台训练了20个epoch的不加时序的模型,但是AP始终为0, 代码结构丝毫未变,可以请教一下什么样的配置可以复现论文的结果?

Mandylove1993 commented 1 year ago

优化器的问题,,但是训练还是不稳定,需要resume才行

LiuXuan66 commented 10 months ago

优化器的问题,,但是训练还是不稳定,需要resume才行

请问能详细说明优化器的具体问题以及如何纠正它们吗

cmj18 commented 7 months ago

请教下,resume之后就能复现精度么