KeyError: 'ExpandBackward'

[lastlaugher]

I tried "python pytorch2caffe" and I've got below error message.

Traceback (most recent call last):
File "pytorch2caffe.py", line 408, in
os.path.join(output_dir, 'inception_v3-pytorch2caffe.caffemodel'))
File "pytorch2caffe.py", line 37, in pytorch2caffe
net_info = pytorch2prototxt(input_var, output_var)
File "pytorch2caffe.py", line 328, in pytorch2prototxt
add_layer(output_var.grad_fn)
File "pytorch2caffe.py", line 154, in add_layer
top_name = add_layer(u[0])
File "pytorch2caffe.py", line 168, in add_layer
layer['type'] = layer_dict[parent_type]
KeyError: 'ExpandBackward'

My pytorch version is the latest one, http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp27-cp27mu-linux_x86_64.whl

[xuzijian]

I have met same problems. I guess the conversion code has not support 'ExpandBackward' currently.

[jnulzl]

met same problems too!

[wangxc2006]

same problem， can anybody give some tips？

[lzhaozi]

met same problems too!

[autocyz]

Let pytorch=0.2, torchvision=0.1.8, the problem will be solved!
I don't kown why, maybe newer pytorch use 'ExpandBackward' to name layer, but older pytorch do not use. And the layer_dict in pytorch2caffe.py do not have 'ExpandBackward'

[wyfdodogod]

does that mean any model trained in torch version 3 and above need to be retrained in torch 2 and then safe to be converted to caffe

[autocyz]

@wyfdodogod I think you needn't retrain in torch 2, you should understand the meaning of ExpandBackward and add this key to layer_dict , then you should add some operater in function pytorch2prototxt, just like elif parent_type == 'BatchNormBackward':

[wyfdodogod]

@autocyz ty for the explaination, as you mentioned using torch 0.2 and torchvision 0.1.8, i solved the problem more specifically torch0.2 post1 using post2 and 3 might raise error causing python crash at the time caffe is imported.

[ys198918]

@wangxc2006 did you retrain the model under torch 0.2?

[elepherai]

Any plan to support 0.4.x ?

[dzyjjpy]

@autocyz @longcw met same issue when running demo.py.
when i try to install 0,2.0 by pip install torch==0.2.0 or pip isntall torch-0.2.0.post********.whl
(I try install it based CUDA 9.2 or CUDA 8.0)
errors displays:
is not a supported wheel on this platform
so strange. dose anyone run demo,py sucessfully on torch==0.3.1 and torchvision==0.2.0
I trained model on the env above. thanks

[konioyxgq]

@autocyz Like you said, "should understand the meaning of ExpandBackward and add this key to layer_dict , then you should add some operater in function pytorch2prototxt, just like elif parent_type == 'BatchNormBackward':"
But i don't know ‘ExpandBackward’ means. So how should it be modified?