VT SailBOT 2020 Sensor Evaluation Repository

Welcome! This repository stores the VT SailBOT 2020 team's code for evaluating our sensor systems, which includes tasks like sorting images into classes for testing (BUOY_PRESENT or BUOY_NOT_PRESENT) and then identifying where in the BUOY_PRESENT images the buoy actually is.

Assumptions

These tools are mainly used for processing image datasets used for testing the sensors. Each tool assumes that testing images are stored in the following directory structure:

TEST_DATASET/ (the top level directory)
    [various directories where raw images are stored after being captured]
    BUOY_PRESENT/       (where images that _have_ a buoy are stored)
    BUOY_NOT_PRESENT/   (where images that _don't have_ a buoy are stored)
    BUOY_MASK_IMAGES/   (where mask images for images from BUOY_PRESENT are stored)

Workflow:

You don't need to place any of the explicitly listed subdirectories in TEST_DATASET/ yourself - the scripts in this repo will make them for you.

1. Take images. Place them into the main TEST_DATASET/ directory under whatever subdirectory names you want (_except the ones explicitly listed above). The single_camera_image_capture.py script from the sailbot-20 repo will default to storing these images in folders based on the date they were taken, such as 2019-11-14_10-40/.

2. Sort/Classify images. Use run_sorting.py to classify each image, placing them into the BUOY_PRESENT or BUOY_NOT_PRESENT directories. These directories will be automatically created if they don't exist.

3. Identify buoy regions ("truth" the images). Use run_truthing.py to identify where the buoy is in each image from BUOY_PRESENT/. The script will create masks according to your selections and place them into BUOY_MASK_IMAGES/, which is automatically created if it doesn't exist.

Details: run_sorting.py

Instructions

The run_sorting.py tool helps users quickly classify images as either (1) containing or (2) not containing a buoy. It displays images one after another on the screen, waiting for the user to:

• Press 'y', indicating the image does have a buoy,
• Press 'n', indicating the image does not have a buoy, or
• Press 'q', quitting the program.

The tool will sort the images into BUOY_PRESENT/ or BUOY_NOT_PRESENT/ depending on your choice.

Use the following command line arguments.

python3 run_sorting.py <path_to_test_dataset_directory>

Details: run_truthing.py

Instructions

The run_truthing.py tool should be run after run_sorting.py, and helps users quickly identify which part of an image contains a buoy. It displays images from BUOY_PRESENT/ one after another, waiting for the user to:

• (1) Click and then (2) drag a circle to represent where the buoy is, displayed as a red circle on the screen,
• Use the h/j/k/l keys to move the created circle left/down/up/right respectively (VIM bindings, if you're familiar with them),
• Use the n/m keys to decrease/increase the size of the circle, respectively,
• Use the space key to confirm a selection and have the program save your selection as an image mask, moving on to the next image, or
• Use the q key to save progress and quit.

Description

The tool itself has two windows:

1. A main window to view the entire image, and
2. A closeup window to view the small area around the cursor in more detail (which is extremely useful when selecting small regions).

The size of the closeup window is based on the closeup_radius command-line argument, which is measured in pixels (recommended value: 20)

Mask files are stored as images with one layer of depth (0..255) in which every pixel is black (0) except those in the user-defined selection, which are white (255).

Each mask file is stored at BUOY_MASK_IMAGES/mask.[name_and_ext_of_original_img].

Use the following command line arguments.

python3 run_truthing.py <path_to_test_dataset_directory> <closeup_radius>