Code for the paper "Deep Learning Pipeline for Automated Visual Moth Monitoring: Insect Localization and Species Classification"
check out the Moth-Scanner Repo
This repository requires cupy>=7.8,<8.0
, but unfortunately, the wheel version supporting CUDA 11 (required for NVIDIA RTX 3XXX GPUs) is only compiled for python3.7 (https://pypi.org/project/cupy-cuda110/7.8.0/).
Hence, we need to install cupy from pip, but before doing so, we have to install the cudatoolkit
package containing nvcc
:
conda create -n moth_scanner python~=3.9.0 mpi4py cython~=0.28
conda activate moth_scanner
conda install -c nvidia -c conda-forge cudnn nccl cutensor cub cuda-thrust \
cudatoolkit~=11.7.0 cudatoolkit-dev~=11.7.0 numpy~=1.23.0
To be sure, check with which nvcc
whether the CUDA-Compiler is successfully installed in your conda environment.
Now, you can install the dependencies with
pip install -r requirements.txt
Run the following to validate the cupy
installation:
python -c "import cupy as cp; cp.show_config(); print(cp.zeros(8) + 2)"
All scripts are located in the scripts
folder.
You are welcome to use our code in your research! If you do so please cite it as:
@article{korsch2023deep,
title={Deep learning pipeline for automated visual moth monitoring: insect localization and species classification},
author={Korsch, Dimitri and Bodesheim, Paul and Denzler, Joachim},
journal={arXiv preprint arXiv:2307.15427},
year={2023}
}
This work is licensed under a GNU Affero General Public License.