This is the implementation of our work - 'Learning Deep Representations and Detection of Docking Stations Using Underwater Imaging'. ##Prerequisite: Programs have been tested on Matlab 2016a. Matconvnet is necessary to run this work. ##Run the code: 1. Test * Please run './Train_and_Test_program/my_detection_test_demo.m'. * Detection results will be saved in './test_samples/results'.
2. Data augmentation
* Please run './Docking_dataaug_program_upload/main.m'.
* Select the datasets through the popping dialogue box.
* The images after data augmentation will be saved in the same directory as the datasets.
* Three files can be obtained after augmentation: proposals_train.mat,proposals_test.mat and docking_imdb.mat. 'Proposals_train.mat' and 'proposals_test.mat' are descriptions of proposals of training and test sets respectively. 'Docking_imdb.mat' describes datasets in the form of imdb.
3. Training
* When training is necessary, please first finish step 2 data augmentation.
* Please copy docking_imdb.mat obtained by augmentation to './data/'. Copy proposals_train.mat and proposals_test.mat to './data/SSW/'.
* Please run './Train_and_Test_program/train.m'.
* The model after training is named by 'net-deployed.mat' and saved in './data/'