AssertionError:Reusage of outer blob name y_grad in Do op

[tangdong1994]

anyone Who has encountered this problem

[orionr]

@tangdong1994 do you still see this issue? Thanks.

[robertparley]

I also stop at here .
Confusing......

[orionr]

@robertparley or @tangdong1994, which tutorial were you running when you saw this? Thanks.

[robertparley]

At first, I am try to do some inference with a detectron model. But I need load the pretrain data with C++.
Now I can convert the detectron model to the caffe2 model(init_net.pb & pred.pb)
So what I want to learn exactly is how to load the caffe2 model using C++, and do some predictions.
Oh,It's a heart-broken work for me..........

[robertparley]

I read the .prototxt file and found all the types of layers in this networks.
As follow:
Conv
Relu
MaxPool
Sum
ResizeNearest
Sigmoid
GenerateProposals
CollectAndDistributeFpnRpnProposals
RoIAlign
Concat
BatchPermutation
FC
Softmax
BBoxTransform
BoxWithNMSLimit
I don't know whether they are supported by caffe2 now.

[stackyjoe]

I'm getting this too. It's the Control_Ops tutorial, second to last runnable cell. I'm on Ubuntu 19.04, I built caffe2 with GPU support. I got the RNN Shakespeare example working(and nvidia-smi confirmed it was using some GPU acceleration), so I think I may have made a mistake during the build process but I'm not entirely sure where.