pytorch-semantic-segmentation

Semantic Image Segmentation using Pytorch

This repository is about the full process from data preprocessing to the training to the evaluation of the models.

Detailed descriptions of my code can be found on my articles on Medium:

• How to Augment Images for Semantic Segmentation

  https://medium.com/@stefan.herdy/how-to-augment-images-for-semantic-segmentation-2d7df97544de?source=friends_link&sk=3978b373c7192775e7acaed4ce357ad9

• Pytorch Semantic Segmentation

  https://medium.com/@stefan.herdy/pytorch-semantic-image-segmentation-b726589662e3?source=friends_link&sk=296c6d5a19044cef272270c7bd39f814

• How to Evaluate Semantic Segmantation Models

  https://medium.com/@stefan.herdy/how-to-evaluate-semantic-segmantation-models-cd2539673701?source=friends_link&sk=94b359ecbb04c3413b3a6d82d42ea21c

• How to Download Labelbox Image Annotations

  https://medium.com/@stefan.herdy/how-to-export-labelbox-annotations-eedb8cb4f217?source=friends_link&sk=fb57499999b624245c4d4713f977a73c

  File: download_mask.py

• How to Upload Labelbox Image Annotations to Perform Active Learning

  https://medium.com/@stefan.herdy/active-learning-with-labelbox-how-to-upload-your-label-masks-99feef86017b?source=friends_link&sk=6525eb8fd4ebfe9b317ddc17247bcd2a

  File: upload_mask.py

To perform further analysis (Curvature Computation) of the segmentation masks you can read the following article:

• Compute the Curvature of a Binary Mask in Python

  https://medium.com/@stefan.herdy/compute-the-curvature-of-a-binary-mask-in-python-5087a88c6288?source=friends_link&sk=7e1bf51044644b678d75e82d310357af

Installation

$ git clone https://github.com/stefanherdy/pytorch-semantic-segmentation.git

Usage

- To train the semantic segmentation model run train.py with "python train.py".
    You can specify the following parameters:
    --batch_size, type=int, default=8, help="Batch Size"
    --learnrate, type=int, default=0.0001, help='learn rate of optimizer'
    --optimizer, choices=['sgd', 'adam'], default='adam'
    --eval_every, type=int, default=1, help="Epochs between evaluation"
    --print_every, type=int, default=1, help="Epochs between print"
    --ckpt_every, type=int, default=20, help="Epochs between checkpoint save"
    --num_classes, type=int, default=8, help="Number of classes"
    --project, choices=['project_1', 'project_2', 'project_3'], default='project_1'
    --resize, type=int, default=512, help="Size of images for resizing"
    --random_crop_size, type=int, default=256, help="Size of random crops. Must be smaller than resized images."

    Example usage:
    "python train.py --eval_every 10 --learnrate 0.00001 --batch_size 16

    
- To evaluate the model run evaluate_model.py with "python evaluate_model.py".
    You can specify the following parameters:
    --num_classes, type=int, default=8, help="Number of classes"
    --batch_size, type=int, default=8, help="Batch Size"
    --project, choices=['project_1', 'project_2', 'project_3'], default='project_1'
    --resize, type=int, default=512, help="Size of images for resizing"
    --random_crop_size, type=int, default=256, help="Size of random crops. Must be smaller than resized images."
    --plot_results, type=bool, default=False, help="Set to True if you want to plot your results. Set False to save results as images"

    Example usage:
    "python evaluate_model.py --batch_size 16

    Make sure you performed the training before, so that the models can be loaded for evaluation.
    Parameters like num_classes, random_crop_size and resize should match your training settings.

License

This project is licensed under the MIT License. ©️ 2023 Stefan Herdy