Skip to content
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Go to file
Cannot retrieve contributors at this time

Installing DensePose

The DensePose-RCNN system is implemented within the detectron framework. This document is based on the Detectron installation instructions, for troubleshooting please refer to the detectron installation document.


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


  • 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.


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 $COCOAPI
# 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 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.


Clone the Densepose repository:

# DENSEPOSE=/path/to/clone/densepose
git clone $DENSEPOSE

Install Python dependencies:

pip install -r $DENSEPOSE/requirements.txt

Set up Python modules:

cd $DENSEPOSE && make

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

python2 $DENSEPOSE/detectron/tests/

Build the custom operators library:

cd $DENSEPOSE && make ops

Check that the custom operator tests pass:

python2 $DENSEPOSE/detectron/tests/

Fetch DensePose data.

Get necessary files to run, train and evaluate DensePose.

cd $DENSEPOSE/DensePoseData

For training, download the DensePose-COCO dataset:


For evaluation, get the necessary files:


Setting-up the COCO dataset.

Create a symlink for the COCO dataset in your datasets/data folder.

ln -s /path/to/coco $DENSEPOSE/detectron/datasets/data/coco

Create symlinks for the DensePose-COCO annotations

ln -s $DENSEPOSE/DensePoseData/DensePose_COCO/densepose_coco_2014_minival.json $DENSEPOSE/detectron/datasets/data/coco/annotations/
ln -s $DENSEPOSE/DensePoseData/DensePose_COCO/densepose_coco_2014_train.json $DENSEPOSE/detectron/datasets/data/coco/annotations/
ln -s $DENSEPOSE/DensePoseData/DensePose_COCO/densepose_coco_2014_valminusminival.json $DENSEPOSE/detectron/datasets/data/coco/annotations/

Your local COCO dataset copy at /path/to/coco should have the following directory structure:

|_ coco_train2014
|  |_ <im-1-name>.jpg
|  |_ ...
|  |_ <im-N-name>.jpg
|_ coco_val2014
|_ ...
|_ annotations
   |_ instances_train2014.json
   |_ ...

Docker Image

We provide a Dockerfile that you can use to build a Densepose image on top of a Caffe2 image that satisfies the requirements outlined at the top. If you would like to use a Caffe2 image different from the one we use by default, please make sure that it includes the Detectron module.

Build the image:

cd $DENSEPOSE/docker
docker build -t densepose:c2-cuda9-cudnn7 .

Run the image (e.g. for BatchPermutationOp test):

nvidia-docker run --rm -it densepose:c2-cuda9-cudnn7 python2 detectron/tests/

To run inference in a docker container based on the prepared docker image, one would need to make all the required data accessible within the container. For that one should first follow the steps described in Fetch DensePose Data section. One could also prefetch all the necessary weights files used for training / inference. Then one should start a container with the host DensePoseData and COCO directories mounted:

nvidia-docker run -v $DENSEPOSE/DensePoseData:/denseposedata -v /path/to/coco:/coco -it densepose:c2-cuda9-cudnn7 bash

Within the container one needs to replace the local DensePoseData directory with the host one:

mv /densepose/DensePoseData /densepose/DensePoseDataLocal
ln -s /denseposedata DensePoseData

and perform steps described in COCO dataset setup:

ln -s /coco /densepose/detectron/datasets/data/coco
ln -s /densepose/DensePoseData/DensePose_COCO/densepose_coco_2014_minival.json /densepose/detectron/datasets/data/coco/annotations/
ln -s /densepose/DensePoseData/DensePose_COCO/densepose_coco_2014_train.json /densepose/detectron/datasets/data/coco/annotations/
ln -s /densepose/DensePoseData/DensePose_COCO/densepose_coco_2014_valminusminival.json /densepose/detectron/datasets/data/coco/annotations/

exit the container and commit the change

docker commit $(docker ps --last 1 -q) densepose:c2-cuda9-cudnn7-wdata

The new image can be used to run inference / training. However, one needs to remember to mount DensePoseData and COCO directories:

nvidia-docker run --rm -v $DENSEPOSE/DensePoseData:/denseposedata -v /path/to/coco:/coco -it densepose:c2-cuda9-cudnn7-wdata <inference_or_training_command>