INSTALL.md

Installation

The code was tested on Ubuntu 18.04, with Anaconda Python 3.6 and PyTorch v1.1 with CUDA 10.0. NVIDIA GPUs are needed for both training and testing. After install Anaconda:

0. [Optional but recommended] create a new conda environment.

   conda create --name CenterNet python=3.6
   

   And activate the environment.

   conda activate CenterNet
   

1. Install pytorch1.1:

   conda install pytorch=1.1 torchvision cuda100 -c pytorch
   

   And disable cudnn batch normalization(Due to this issue).

   For other pytorch version, you can manually open torch/nn/functional.py and find the line with torch.batch_norm and replace the torch.backends.cudnn.enabled with False. We observed slight worse training results without doing so.

2. Install [COCOAPI]:

   pip install pycocotools
   

3. Clone this repo:

   CenterNet_ROOT=/path/to/clone/CenterNet
   git clone https://github.com/xingyizhou/CenterNet $CenterNet_ROOT
   

4. Install the requirements

   pip install -r requirements.txt
   

5. Compile deformable convolutional (from DCNv2).

   cd $CenterNet_ROOT/src/lib/models/networks/DCNv2
   conda install conda
   conda install -c conda-forge cudatoolkit-dev=10.0
   python setup.py build develop
   

6. [Optional, only required if you are using extremenet or multi-scale testing] Compile NMS if your want to use multi-scale testing or test ExtremeNet.

   cd $CenterNet_ROOT/src/lib/external
   python setup.py build_ext --inplace
   

7. Download pertained models for detection or pose estimation and move them to $CenterNet_ROOT/models/. More models can be found in Model zoo.