Installing DensePose

Because installing DensePose is hard, I provided a simple installation guid based on DensePose Installation and the provided Dockerfile.

Requirements:

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

Notes:

We are using CUDA 9.0 with cuDNN 7.1 for this tutorial

Getting Started

Installing a fresh Ubuntu 16.04
1. updating ubuntu:
```
sudo apt-get update
sudo apt-get upgrade
```
2. reboot for booting up to the last kernel version
```
reboot
```
We need to have linux kernel 4.4.0 since it's the only kernel that is supported for CUDA 9.0 and ubuntu repositories only provide linux source for that kernel. For Ubuntu 16.04 LTS on x86-64, both the HWE kernel (4.13.x for 16.04.4) and the server LTS kernel (4.4.x) are supported in CUDA 9.2. Visit https://wiki.ubuntu.com/Kernel/Support for more information.
```
sudo apt-get install build-essential
sudo apt-get install linux-image-extra-virtual
sudo apt-get install linux-source
reboot
```
make sure you are booting into the right kernel version ( 4.4.0 ). since we loaded kernel 4.4.0 we should see the output of uname something like this:
```
uname -r
```
```
4.4.0-128-generic
```
then we should install the kernel headers
```
sudo apt-get source linux-image-$(uname -r)
sudo apt-get install linux-headers-$(uname -r)
```
Downloading cuda 9.0 runfile
Download cudnn7 ( possibly 7.1 )
install cuda 9.0:
1. Create a file at /etc/modprobe.d/blacklist-nouveau.conf with the following contents
```
blacklist nouveau
options nouveau modeset=0
```
2. Regenerate the kernel initramfs:
```
sudo update-initramfs -u
```
3. reboot to runlevel 3 adding character 3 to the end of the grub bootloader at boot time will get you to the runlevel 3
4. go to you cuda directory
```
sudo sh ./cuda-installer
```

installing cudnn 7 using nvidia documentation

it's quite easy using the tar file

tar -xzvf cudnn-9.0-linux-x64-v7.tgz

Copy the following files into the CUDA Toolkit directory.

sudo cp cuda/include/cudnn.h      /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn*      /usr/local/cuda/lib64
sudo chmod a+r  /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

Next we have to download Nvidia NCCL for ubuntu
1. installing nccl using this guide
We should also download anaconda to install caffe2
1. Download Python 2.7 version from this link
```
chmod +x Anaconda2-5.2.0-Linux-x86_64.sh
./Anaconda2-5.2.0-Linux-x86_64.sh
```
2. The installation will ask you for some location but for testing the default ones will work
3. It will install anaconda to your home directory
4. We assume that anaconda is installed in our home directory and the path looks like this:
```
/home/terra/anaconda2  
```
  adding the path to our environment path
```
export PATH="/home/terra/anaconda2/bin${PATH:+:${PATH}}"
export PATH="/usr/local/cuda/bin${PATH:+:${PATH}}"
export LD_LIBRARY_PATH=/usr/local/cuda/lib64\
                        ${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
```

Installing caffe2

By running this command conda will try to install caffe2 with cuda support.
```
conda install -c caffe2 caffe2-cuda9.0-cudnn7
```

Verifing caffe2 installation:

python2 -c 'from caffe2.python import     workspace; print    (workspace.NumCudaDevices())'

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.

Finally Installing Densepose

Clone the Densepose repository:

# DENSEPOSE=/path/to/clone/densepose
git clone https://github.com/facebookresearch/densepose $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/test_spatial_narrow_as_op.py

Build the custom operators library:

cd $DENSEPOSE && make ops

Check that the custom operator tests pass:

python2 $DENSEPOSE/detectron/tests/test_zero_even_op.py

Fetch DensePose data.

Get necessary files to run, train and evaluate DensePose.

cd $DENSEPOSE/DensePoseData
bash get_densepose_uv.sh

For training, download the DensePose-COCO dataset:

bash get_DensePose_COCO.sh

For evaluation, get the necessary files:

bash get_eval_data.sh

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