/MDvsFA

PyTorch implementation of ICCV2019 paper Miss Detection vs. False Alarm: Adversarial Learing for Small Object Segmentation in Infrared Images.

Primary LanguagePython

MDvsFA

PyTorch implementation of ICCV2019 paper Miss Detection vs. False Alarm: Adversarial Learing for Small Object Segmentation in Infrared Images.

Guide

  1. Creating the following folders:

    • training_results: this folder is to contain all the images of evaluation phases, to visualize the performance of model.
    • test_results: this folder is to contain the images during test phases.
    • logs: this folder is to contain all logs during training.
    • saved_models: to save the weight after each epoch.

    The following command is to create fodler under the root of repository:

    mkdir training_results test_results logs saved_models
  2. Dataset: The official implementation offers the dataset, the structure has to be:

    root
        data
            test_gt
            test_ort
            training
    
  3. Using following command to train:

    python train.py

    all the training parameters have default values.

  4. Using following command to test:

    python test.py