CPU Memory about unet_mxnet2ncnn

[xuduo35]

How much batch size did you set?

[Tomhouxin]

16

[xuduo35]

16 is not big. But you can try to decrease it, or input image size. I don't remember if I met with this issue or not. But I'm using Python3.5 and mxnet-cu80. Which line of trainunet.py cause this crash?

[Tomhouxin]

I am also using Python3.5 on ubuntu14.04, not GPU. I can't remember which line. I'll let you know tomorrow.
I wonder this is running on android

[xuduo35]

ncnn can run on android.

[Tomhouxin]

line 73，like ‘curr_loss = F.mean(loss).asscalar()’. ncnn can run on android, but trainunet ?

[xuduo35]

Seems your GPU's memory is not enough. You can try use small filter numbers. trainunet is only for training. Inference don't need these python scripts, only models.