search

facebookarchive / tutorials

Caffe2 Tutorials
Apache License 2.0
123 stars 61 forks source link

AssertionError:Reusage of outer blob name y_grad in Do op #13

Open tangdong1994 opened 6 years ago

tangdong1994 commented 6 years ago

anyone Who has encountered this problem wx20181005-195748 wx20181005-195811 wx20181005-195829

orionr commented 5 years ago

@tangdong1994 do you still see this issue? Thanks.

robertparley commented 5 years ago

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

orionr commented 5 years ago

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

robertparley commented 5 years ago

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 commented 5 years ago

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 commented 5 years ago

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.