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