Pipeline for semantic segmentation

This project has been developed during my internship at LISSI. The objective was to create a pipeline for semantic segmentation with python and pytorch.

1. Documentation

You can read the documentation here.

2. How to use the pipeline

git clone https://github.com/MalondaClement/pipeline.git

2.1 Training

2.1.1 Parameters for training

Before starting training it's possible to change training parameters using ARGS constructor in train.py script.

Parameters of the constructor :

• Models name,
• Dataset path,
• Number of classes,
• Type of labels the dataset used ("label", "csv" or "json") (depends of the dataset),
• Batch size,
• Number of epoch.

2.1.2 Start training

python3 train.py

1.3 Save directory

pipeline
└── save
    ├── model_name
    │   ├── yyyy-mm-dd-id
    │   │   ├── best_weights.pth.tar
    │   │   ├── checkpoint.pth.tar
    │   │   └── learning_curves.png
    │   └── yyyy-mm-dd-id
    │   │   ├── best_weights.pth.tar
    │   │   ├── checkpoint.pth.tar
    │   │   └── learning_curves.png
    └── model_name
        └── yyyy-mm-dd-id
            ├── best_weights.pth.tar
            ├── checkpoint.pth.tar
            └── learning_curves.png

2.2. Evaluation

python3 evaluation.py

2.3. Inference

python3 inference.py