Incubator-mxnet: AttributeError: 'module' object has no attribute 'Proposal' when running RCNN train_end2end.py

[krosaen]

  File "train_end2end.py", line 184, in <module>
    main()
  File "train_end2end.py", line 181, in main
    lr=args.lr, lr_step=args.lr_step)
  File "train_end2end.py", line 30, in train_net
    sym = eval('get_' + args.network + '_train')(num_classes=config.NUM_CLASSES, num_anchors=config.NUM_ANCHORS)
  File "/root/mxnet/example/rcnn/rcnn/symbol/symbol_vgg.py", line 359, in get_vgg_train
    rois = mx.symbol.Proposal(
AttributeError: 'module' object has no attribute 'Proposal'

Built from source as of commit 2916b4e7266f596639c415411238d5a01555fcf4 (big rcnn change today).

Here is my Dockerfile:

FROM nvidia/cuda:8.0-cudnn5-devel

# mxnet
#
# note: we build it ourselves instead of using FROM kaixhin/cuda-mxnet:8.0 to control
# the version of mxnet we are building. Bleeding edge master is not always fun.
#

# Install git, wget and other dependencies
RUN apt-get update && apt-get install -y \
  git \
  libopenblas-dev \
  libopencv-dev \
  python-dev \
  python-numpy \
  python-setuptools \
  wget

# Clone MXNet repo and move into it
RUN cd /root && git clone --recursive https://github.com/dmlc/mxnet && cd mxnet && \
# use a fixed stable point (key fixes to rcnn on 1/19/2017 https://github.com/dmlc/mxnet/pull/4730)
  git checkout 2916b4e7266f596639c415411238d5a01555fcf4 && \
# Copy config.mk
  cp make/config.mk config.mk && \
# Set OpenBLAS
  sed -i 's/USE_BLAS = atlas/USE_BLAS = openblas/g' config.mk && \
# Set CUDA flag
  sed -i 's/USE_CUDA = 0/USE_CUDA = 1/g' config.mk && \
  sed -i 's/USE_CUDA_PATH = NONE/USE_CUDA_PATH = \/usr\/local\/cuda/g' config.mk && \
# Set cuDNN flag
  sed -i 's/USE_CUDNN = 0/USE_CUDNN = 1/g' config.mk && \
# Make
  make -j"$(nproc)"

# Install Python package
RUN cd /root/mxnet/python && python setup.py install


# opencv
#
# Note: we used to build and install opencv3 but that started causing conflicting library issues because
# opencv 2.7 is included via libopencv-dev above. Simply installying the python bindings to the same library
# seems to be sufficient.

RUN apt-get install -y python-opencv

# random other dependencies

RUN apt-get install -y python-scipy \
  python-matplotlib \
  python-pip

RUN pip install easydict cython scikit-image

# cythonisize!?!?!
RUN cd /root/mxnet/example/rcnn && make && cd -

WORKDIR /root/mxnet/example/rcnn

[krosaen]

perhaps @piiswrong or @precedenceguo have insight as this might be related to recently merged https://github.com/dmlc/mxnet/pull/4730

[piiswrong]

You need to compile the extra operators. It's in the README

[krosaen]

thanks @piiswrong, I've been working with RCNN for so many weeks I'd become too accustom to the previous dependencies.

[great-thoughts]

I am getting the same error. how did you fix your problem @krosaen ?

[krosaen]

@great-thoughts you can see how we run this here: https://github.com/umautobots/nn-dockerfiles/tree/master/mxnet-rcnn

if you don't want to use docker, the commands in the dockerfile still provides a reproducible build, the key update is:

sed -i 's/EXTRA_OPERATORS =/EXTRA_OPERATORS = example\/rcnn\/operator/g' config.mk && \

[great-thoughts]

@krosaen The error still persists: Below is exactly what I did. I appreciate if you would help.

    1  cd mxnet
    2  git pull
    3  cd example/rcnn
    4  ls
    5  bash script/additional_deps.sh
    6  pip install matplotlib
    7  bash script/get_voc.sh
    8  bash script/get_pretrained_model.sh
    9  bash script/vgg_voc07.sh 0,1
   10  ls
   11  cd ..
   12  sed -i 's/EXTRA_OPERATORS =/EXTRA_OPERATORS = example\/rcnn\/operator/g' config.mk
   13  make -j"$(nproc)"
   14  cd example
   15  cd rcnn
   16  make
   17  bash script/vgg_voc07.sh 0,1

[great-thoughts]

@piiswrong I followed the readme but still getting an error.

[ijkguo]

If you really don't know how to do this, just copy everything in example/rcnn/operator into src/operator

[krosaen]

@great-thoughts I can't immediately spot the issue, but if you try our docker file, it should work. It's possible there's an issue at the tip of master; our dockerfile builds from v0.9.3 to ensure stability (we'll update it as we test out new releases)

[zht3344]

@great-thoughts I am getting the same error, Did you solve this problem?

[great-thoughts]

No, I have not. @zht3344

[joyivan]

I spot the same error after I cp ~/mxnet/example/rcnn/operator/* ~/mxnet/src/operator

[joyivan]

@great-thoughts do you solve it?

[wengyc]

I also have the same error after set EXTRA_OPERATORS = example/rcnn/operator.Please give me some advices.

[great-thoughts]

Hi @piiswrong , I was wondering if you can help solve this issue since many of us are facing a problem.

[ijkguo]

1. First of all, delete all mxnet in your machine. Make sure import mxnet fails.
2. Get a mxnet, copy example/rcnn/operator/* to src/operator/*.
3. Compile mxnet and install mxnet.

If you still have no attribute Proposal:

1. ls build/src/operator and check if proposal.o is present.
2. If not, you must have missed something in the above 3 steps.
3. If yes, reply here.

[wengyc]

I set the EXTRA_OPERATORS = ~/mxnet/example/rcnn/operator in the file "config.mk" and re-make the mxnet,now it work.

[great-thoughts]

Yes, It is working now for me.

[liangfu]

The source code is more recently (v0.9.5) moved to mxnet/src/operator/contrib, and adding to EXTRA_OPERATORS is no longer necessary. However, we should import Proposal by importing from mxnet.contrib.symbol instead of mxnet.symbol package. Hope this helps!~