yolov4在rk1808s0计算棒的耗时

leok 发表于 2020-6-22 08:46
和骨干网络设计和本算算法复杂度相关。

Did you guys find a solution to this issue? Do we have to remove spp structure from yolov4 backbone and then retrain with entire imagenet to solve it?

leok 发表于 2020-11-13 09:05
已经发布。https://github.com/rockchip-linux/rknn-toolkit/tree/master
同时需升级计算棒固件。 ...

Hey Leok,

Do you know why maxpool in spp structure of yolov4 is taking too much time?
Is there a way to solve this bottleneck, because currently even though mish layer is supported, yolov4 is not worth using due to its huge inference time (like 200ms)