Installation

The Installation step is referenced from CenterNet.

Our experimental environment:

Ubuntu 18.04, Python 3.6.9, PyTorch 1.4, torchvision 0.5.0.

8 NVIDIA 2080ti cuda10.0

Create a new conda environment and activate the environment.
```
conda create --name MOC python=3.6
conda activate MOC
```
Install pytorch1.4:
```
pip install in pytorch.org.
- or you can use conda. 
```
Disable cudnn batch normalization(follow CenterNet).

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.

Clone this repo:

git clone https://github.com/NEUdeep/MOC-Detector-Pytorch1.4

Install the requirements

pip install -r requirments.txt -i https://mirrors.aliyun.com/pypi/simple

Compile deformable convolutional in DLA backbone follow CenterNet.

cd ${MOC_ROOT}/src/network/DCNv2
python setup.py build develop
- or you must check with https://github.com/CharlesShang/DCNv2.git to make sure that is for you.

If you have any problems, please check in CenterNet.