INSTALL.md

Installation

This document contains detailed instructions for installing the necessary dependencies for YOLOv4-pytorch. The instrustions have been tested on an Ubuntu 18.04 system and windows 10. We recommend using the install script if you have not already tried that.

Requirements

• Conda installation with Python 3.6. If not already installed, install from https://www.anaconda.com/distribution/.
• Nvidia GPU.
• CUDA10.0
• CUDNN7.0

Step-by-step instructions

Create and activate a conda environment

conda create --name YOLOv4-pytorch python=3.6
conda activate YOLOv4-pytorch

Install PyTorch

Install PyTorch with cuda10.0

conda install pytorch torchvision cudatoolkit=10.0 -c pytorch

Note:

• It is possible to use any PyTorch supported version of CUDA (not necessarily v10).
• For more details about PyTorch installation, see https://pytorch.org/get-started/previous-versions/.

Install numpy,opencv-python, tqdm, argparse, pickleshare and tensorboardX

pip install -r requirements.txt --user

Install the coco toolkit

If you want to use COCO dataset for training, install the coco python toolkit. You additionally need to install cython to compile the coco toolkit.

conda install cython
pip install pycocotools
or using
pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI
more information please see https://github.com/philferriere/cocoapi