EVA-Space-Center-Data-Generate

Introduction

Time

- A project started from 2019 June.

Goal

- It's aims to predict the position and the pose of an aircraft from a single Moon image.

Data consists of 100,000 moon images from random angles and distance.
80,000 train data；10,000 test data；10,000 valid data.
The label(target/ground truth): c_gamma, c_theta, c_phi, p_gamma, p_theta, p_phi, u_x, u_y, u_z
The global world coordinate is based on Moon center, all vectors are based on the origin of the global world coordinate.
We consider the spherical coordinate in physics view.
The meaning of this 9 parameters list below:
- c_gamma: gamma of the camera position.
- c_theta: theta of the camera position.
- c_phi: phi of the camera position.
- p_gamma: gamma of the optical axis' end point. However, the setting of p_gamma doesn't cause any difference of where cemera look at nor the images, only the direction matters, you can see the experiment below.
- p_theta: theta of the optical axis' end point.
- p_phi: phi of the optical axis' end pount.
- u_x: x componet of camera pose's normal vecter.
- u_y: y componet of camera pose's normal vecter.
- u_z: z componet of camera pose's normal vecter.
The range of this 9 parameters list below:
- c_gamma: [1.74308766628, 1.75292031414] in OpenGL unit --> [1737.3, 1747.1] km, 200m ~ 10,000m above Moon surface.
- c_theta: [0, pi] radian
- c_phi: [0, 2pi] radian
- p_gamma: [0, 1.742887] in OpenGL unit --> [0, 1737.1] km, radius of the Moon.
- p_theta: [0, pi] radian
- p_phi: [0, 2pi] radian
- u_x: [-1, 1] no unit, since the normal vector is normalized.
- u_y: [-1, 1] no unit , since the normal vector is normalized.
- u_z: [-1, 1] no unit, since the normal vector is normalized.

    void gluLookAt(	GLdouble eyeX, GLdouble eyeY, GLdouble eyeZ,
                        GLdouble centerX, GLdouble centerY, GLdouble centerZ,
                        GLdouble upX, GLdouble upY, GLdouble upZ
                   );

gluLookAt Parameters Meaning:
- eyeX, eyeY, eyeZ
  - Specifies the position of the eye point.
- centerX, centerY, centerZ
  - Specifies the position of the reference point.
- upX, upY, upZ
  - Specifies the direction of the up vector.
See more definition of gluLookAt at https://www.khronos.org/registry/OpenGL-Refpages/gl2.1/xhtml/gluLookAt.xml

Enviroment

Anaconda 2
python 3.6
pip install
- opencv-python==3.4.2
- pygame==1.9.4
- PyOpenGL==3.1.3b2

Config

Before generating dataset, you have to adjust config.py.

Set logger
- Which level of logger you want to show.
Dateset name
- Name the dataset.
Units
- Units of the Moon and the openGL world.
Constraints:
- Set your image size, field of view and the range of c_gamma.
- VIEWPORT: The size of the image.
- FOVY: The field of view angle, in degrees, in the y direction.
- Z_NEAR: The distance from the viewer to the near clipping plane (always positive).
- Z_FAR: The distance from the viewer to the far clipping plane (always positive).
- LOWER_BOUND: The lower bound of the c_gamma.
- UPPER_BOUND: The upper bound of the c_gamma.
PATH
- The path you save your dataset.
hyperparameters
- LEVEL_1_INDEX: How many part_1 do you want.
- LEVEL_2_INDEX: How many part_2 do you want.
- IMAGE_INDEX: How many images do you want in one part_2.

Setting up OpenGL Envoriment

set_viewport():
- Set view port.
set_light_property():
- Set light poroperty.
set_filed_of_vision():
- Set FOV.
set_camera_position():
- Set camera position. c_gamma, c_theta, c_phi are sampled with random.uniform in their own range.
set_optical_axis_look_at():
- Set optical axis' end point. p_gamma, p_theta, p_phi are sampled with random.uniform in their own range.
camera_direction():
- Set direction of the camera. u_x, u_y, u_z are sampled with random.uniform between 0~1.

JSON File Format

Each JSON file contains 10,000 images/tagets.
Hierarchy of the directory

Dataset_all_random

Level 1 directory i: 0 ~ 9

i_j.targz

target_i.json

Level 2 directory i_j j: 0 ~ 9

1,000 images in name of "DATASETNAME_XXXXX.png"

target_i.json:

  {
    'image_name': [`c_gamma`, `c_theta`, `c_phi`, `p_gamma`, `p_thet`a, `p_phi`, `u_x`, `u_y`, `u_z`]
    }

Dataset_all_random_2

Level 1 directory i: 0 ~ 9

i_j.targz

target_i.json

Level 2 directory i_j j: 0 ~ 9

1,000 images in name of "DATASETNAME_XXXXX.png"

target_i.json:

  {
    'image_name': {
                     'spherical': [`c_gamma`, `c_theta`, `c_phi`, `p_gamma`, `p_thet`a, `p_phi`, `u_x`, `u_y`, `u_z`],
                     'cartesian': [`c_x`, `c_y`, `c_z`, `p_x`, `p_y`, `p_z`, `u_x`, `u_y`, `u_z`]
                    }
    }

Dataset_six_random (fix Up vector with (0, 1, 0))

Level 1 directory i: 0 ~ 9

i_j.targz

target_i.json

Level 2 directory i_j j: 0 ~ 9

2,000 images in name of "DATASETNAME_XXXXX.png"

target_i.json:

  {
    'image_name': {
                     'spherical': [`c_gamma`, `c_theta`, `c_phi`, `p_gamma`, `p_thet`a, `p_phi`, `u_x`, `u_y`, `u_z`],
                     'cartesian': [`c_x`, `c_y`, `c_z`, `p_x`, `p_y`, `p_z`, `u_x`, `u_y`, `u_z`]
                    }
    }