Dreamgang
Dreamgang
I have got the same problem!
> I have resolved the problem: > change the scope "mobilenet_v2" to "MobilenetV2" 方便加个微信吗?
> I have resolved the problem: > change the scope "mobilenet_v2" to "MobilenetV2" This problem has resolved,but I have got another problem `Assign requires shapes of both tensors to match....
Hey @ @nerox8664 I've updated the pytorch2keras package to the latest, and the errors still exists.
> 在VS里调试运行。你要是直接运行,在运行目录里没有model这个文件夹 谢谢,已经调试通过了。大致测试了一下,发现在windows下4_128x96.jpg的输入,耗时差不多在13ms左右,跟您说的3.5ms,有点差距啊,我不知道问题出在哪? - rnet = 0.5 M, onet = 2.1 M convert cost: 0.133 ms nms cost: 0.776 ms, (27-->6) nms cost: 0.067 ms, (3-->2) nms cost: 0.032 ms,...
> 用release 老师,还有最后一个问题,咱这个模型的mini_face_sizes是多少,是60吗?
> 你没改代码的话是20 好的,谢谢您。
@zuoqing1988 左博,请问在rk3288上,640*480的输入,最小人脸80,检测速度能做到30fps以上吗?谢谢!
> @Dreamgang ZQCNN应该很难,但是把模型转到ncnn或者MNN应该可以 好的,谢谢
when I run the demo on the rk3399,the cost time is about 300ms per image ,