Introduction and usage of Detectron

mask-detectron-post_o.jpg

Installing Detectron

This document covers how to install Detectron, its dependencies (including Caffe2), and the COCO dataset.

  • For general information about Detectron, please see README.md.

Requirements:

  • NVIDIA GPU, Linux, Python2
  • Caffe2, various standard Python packages, and the COCO API; Instructions for installing these dependencies are found below

Notes:

  • Detectron operators currently do not have CPU implementation; a GPU system is required.
  • Detectron has been tested extensively with CUDA 8.0 and cuDNN 6.0.21.

Caffe2

To install Caffe2 with CUDA support, follow the installation instructions from the Caffe2 website. 

If you already have Caffe2 installed, make sure to update your Caffe2 to a version that includes the Detectron module.

Please ensure that your Caffe2 installation was successful before proceeding by running the following commands and checking their output as directed in the comments.

# To check if Caffe2 build was successful
python2 -c 'from caffe2.python import core' 2>/dev/null && echo "Success" || echo "Failure"

# To check if Caffe2 GPU build was successful
# This must print a number > 0 in order to use Detectron
python2 -c 'from caffe2.python import workspace; print(workspace.NumCudaDevices())'

If the caffe2 Python package is not found, you likely need to adjust your PYTHONPATH environment variable to include its location (/path/to/caffe2/build, where build is the Caffe2 CMake build directory).

Other Dependencies

Install the COCO API:

# COCOAPI=/path/to/clone/cocoapi
git clone https://github.com/cocodataset/cocoapi.git $COCOAPI
cd $COCOAPI/PythonAPI
# Install into global site-packages
make install
# Alternatively, if you do not have permissions or prefer
# not to install the COCO API into global site-packages
python2 setup.py install --user

Note that instructions like # COCOAPI=/path/to/install/cocoapi indicate that you should pick a path where you'd like to have the software cloned and then set an environment variable (COCOAPI in this case) accordingly.

Detectron

Clone the Detectron repository:

# DETECTRON=/path/to/clone/detectron
git clone https://github.com/facebookresearch/detectron 
$DETECTRON

Install Python dependencies:

pip install -r $DETECTRON/requirements.txt

Set up Python modules:

cd $DETECTRON && make

Check that Detectron tests pass (e.g. for SpatialNarrowAsOp test):

python2 $DETECTRON/detectron/tests/test_spatial_narrow_as_op.py

That's All You Need for Inference

At this point, you can run inference using pretrained Detectron models.
Take a look at our inference tutorial for an example.
If you want to train models on the COCO dataset, then please continue with the installation instructions.

Datasets

Detectron finds datasets via symlinks from detectron/datasets/data to the actual locations where the dataset images and annotations are stored.
For instructions on how to create symlinks for COCO and other datasets, please see detectron/datasets/data/README.md.
After symlinks have been created, that's all you need to start training models.

Troubleshooting

In case of Caffe2 installation problems, please read the troubleshooting section of the relevant Caffe2 installation instructions first. In the following, we provide additional trouble shooting tips for Caffe2 and Detectron.

Caffe2 Operator Profiling

Caffe2 comes with performance profiling support which you may find useful for benchmarking or debugging your operators (see BatchPermutationOp test for example usage).
Profiling support is not built by default and you can enable it by setting the -DUSE_PROF=ON flag when running Caffe2 CMake.

CMake Cannot Find CUDA and cuDNN

Sometimes CMake has trouble with finding CUDA and cuDNN dirs on your machine.

When building Caffe2, you can point CMake to CUDA and cuDNN dirs by running:

cmake .. \
  # insert your Caffe2 CMake flags here
  -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda/toolkit/dir \
  -DCUDNN_ROOT_DIR=/path/to/cudnn/root/dir

Similarly, when building custom Detectron operators you can use:

cd $DETECTRON
mkdir -p build && cd build
cmake .. \
  -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda/toolkit/dir \
  -DCUDNN_ROOT_DIR=/path/to/cudnn/root/dir
make

Note that you can use the same commands to get CMake to use specific versions of CUDA and cuDNN out of possibly multiple versions installed on your machine.

Protobuf Errors

Caffe2 uses protobuf as its serialization format and requires version 3.2.0 or newer.
If your protobuf version is older, you can build protobuf from Caffe2 protobuf submodule and use that version instead.
To build Caffe2 protobuf submodule:

# CAFFE2=/path/to/caffe2
cd $CAFFE2/third_party/protobuf/cmake
mkdir -p build && cd build
cmake .. \
  -DCMAKE_INSTALL_PREFIX=$HOME/c2_tp_protobuf \
  -Dprotobuf_BUILD_TESTS=OFF \
  -DCMAKE_CXX_FLAGS="-fPIC"
make install

To point Caffe2 CMake to the newly built protobuf:

cmake .. \
  # insert your Caffe2 CMake flags here
  -DPROTOBUF_PROTOC_EXECUTABLE=$HOME/c2_tp_protobuf/bin/protoc \
  -DPROTOBUF_INCLUDE_DIR=$HOME/c2_tp_protobuf/include \
  -DPROTOBUF_LIBRARY=$HOME/c2_tp_protobuf/lib64/libprotobuf.a

You may also experience problems with protobuf if you have both system and anaconda packages installed.
This could lead to problems as the versions could be mixed at compile time or at runtime.
This issue can also be overcome by following the commands from above.

AUTHOR

READ NEXT

Boostlog is an online community for developers
who want to share ideas and grow each other.

Delete an article

Deleted articles are gone forever. Are you sure?