Jie Zhang

This error depends on your system platform.

sorry, I have no idea. I didn't test it on arm linux.

Yes, It's a bug. Checkout branch [`fix-create-layer`](https://github.com/luoyetx/mini-caffe/tree/fix-create-layer). I haven't merge it yet. The bug is from cudnn layers, when you build with `-DUSE_CUDNN=ON`.

Sorry, I merged the branch. The code is now in master branch.

What does `a crash happened` mean? segment fault? Did you initialize the network parameters?

@wuqiangch what's your cuda version?

use `net->blob_by_name()` function to get data pointer from network internal blobs. You can refer to the [exampes](https://github.com/luoyetx/mini-caffe/tree/master/example).

I have no experience with TensorFlow. Did you train the model on MNIST or other dataset. It's weird to see loss value to be that low, how's the test loss?...

take a look at the derivatives. here is [the latex version](https://luoyetx.github.io/2017/05/derivatives-for-lsoftmax/#derivatives-for-lsoftmax)

Yes. The first item is the original fc derivatives.