Free cookie consent management tool by TermsFeed
Back Arrow
Go To Github

Sophie Wildeboer

Weeds detections using Transfer Learning

05/06/2023

Downloads Contributors Forks Stargazers Issues License

workflow

Table Of Contents

Content

This repository covers the dataset creation/preparation, model training and evaluation code and model weights for transfer learning YOLOv7 on a agricultural-specific pretraining dataset. Compared to COCO pretraining. All the code and methods are found in their respective notebooks in the notebooks folder. Additionally, pretrained YOLOv7 model weights on are found here: MEGAWEEDS_YOLOv7 Finally the MEGAWEEDS dataset can be found on Zenodo over here: MEGAWEEDS dataset.

Installation

There are 2 different ways to get started: 1. Python installation or, 2. A precompiled docker-container

1: Anaconda

Get your preferred Python installation method (Pip, Conda, miniforge, etc.) and install ultralytics, pytorch and jupyterlab

2: Docker

Not wanting to deal with anaconda and environment.yamls? Run everything in a docker container:

  1. Install Docker
  2. Start Docker
  3. Start the Jupyter-lab docker from your command-line with docker installed and ready: this pulls the most recent ddal with all packages and libraries installed.
    docker run --rm -it -p 8888:8888 --shm-size=5gb -v /directory/to/this/repo/on/local/machine:/home/jovyan docker/container:tag

Or using GPUs: Install Docker with NVIDIA runtime and run a gpu-enabled docker container:

    docker run --rm -it --runtime=nvidia -p 8888:8888 --gpus 1 --shm-size=5gb -v /directory/to/this/repo/on/local/machine:/home/jovyan docker/container:tag

Acknowledgements

This project is funded by the European Union, grant ID 101060643.

Authors