/DeepWeeds_ComputerVision

Primary LanguageJupyter Notebook

DeepWeeds : A Multiclass Weed Species Classification Using Deep Learning


Summary

The project is a multiclass weed species classification model built using deep learning algorithms As a replication of the results of this paper DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning.

Dataset

  • The dataset used in this project is the DeepWeed provided by the paper Authors and made available throught TensorFlow Datasets.
  • deep_weeds DataSet consists of 17,509 labelled images of eight nationally significant weed species native to eight locations across northern Australia.

Model Architecture

  • Pretrained ResNet50 with Imagenet weights.
  • Pretrained InceptionV3 with Imagenet weights.
  • Adam Optimizer with Expotetial Learning Rate decay.

Dependencies

  • Python 3.7 64bit
  • Windows 10 64 bit / version 1909
  • Nvidia Geforce RTX 2060 laptop
  • cuda_11.0.2_451.48_win10
  • cudnn-11.0-windows-x64-v8.0.1.13
  • Bazel 3.1.0
  • Jupyter Notebook or Jupyter Lab

Recommended

Note that is better to create a new virtual environment and install requirements in it to isolate your project from your default environment and to avoid all problems that caused by Package versions conflict.

  • The first think is to install virtualenv.

    pip install virtualenv

  • Clone the project.

    cd deepweeds_computervision

  • Now the folder venv is you main environment and you can name it what are every you want. and all the packages you will install it will store there.

    virtualenv \venv

  • To activate your virtual environment.

    venv\Scripts\activate.bat

  • Now you can install any python packages you want in the command line that you are activating the virtual environment and they will be isolated.

  • To exit the virtual environment.

    deactivate

  • For more information check here.

Requirements

  • All requirements that you will need with its version it's exist in requirements.txt so you need just to run this command to install it all :

    !pip install -r requirements.txt

  • Download the pre-trained weights here then place them to \Checkpoint folder.

How to run it

After Installing all The requirememnets and activating your virtual environment Start Jupyter Lab or Jupyter Notebook Commadsjupyter lab or jupyter Notebook .

jupyter lab

Then open the browser and tap http://localhost:8888/lab and do the heck you want with it. If you use jupyter Notebook you know what to do.

Training the model

  • You can either download the pretrained weights or Run the re-train the model.
  • find pretrained weights of the model here

For monitoring reasons, we are using the Livelossplot that is shown in the training phase and also we use the tensorboard.

To observe the accuracy during the training run this command to launch the tensorboard servers in the localhost.

tensorboard --logdir logs/fit

Authors

Thank you.