Train data for NeuralTexture with PyOpengl

This repository generates training data (UV) for NeuralTexture using OpenGL.

---

Prerequisite

1. Python v3.x
2. PyOpenGL v3.1.5
3. PyGLFW v1.11.0
4. Pyrr v0.10.3
5. Numpy v1.15 or newer
6. PIL v6.2.0 or newer

---

Usage

You may need to modify python code to have different result. All results will be saved in result folder.

Get UV data

In main.py on line 13 change get_uv to True to get UV data. Otherwise you can just see mesh with some texture.

Save data

To save data, in main.py set save_data on line 14 as True. Depends of get_uv in main.py you can save screenshot (as png file) if you set get_uv equal True, otherwise it will be numpy array (as npy file)

---

Transformation used on loading Mesh (3D object)

Examples of transformation: Create identity matrix and variables that will be store transformation such as translation and rotate.

matrix = pyrr.Matrix44.identity()
translation = pyrr.Vector3()
rotate = pyrr.quaternion.create()

Move mesh relative to the x, y, z axis. After that you need multiply general matrix to this transformation (further we do this with others transformations).

translation += [0.0, 0.0, -0.0]
translation_matrix = pyrr.matrix44.create_from_translation(translation)
matrix *= translation_matrix

Rotation. Here example across axis y.

rotation = pyrr.quaternion.create_from_y_rotation(np.pi / 2.0 * glfw.get_time())
rotate = pyrr.quaternion.cross(rotation, rotate)
rotate = pyrr.matrix44.create_from_quaternion(rotate)
matrix *= rotate

Scale. Vector3(x, y, z) means that scale will be used on certain axis with certain coefficient.

scale_matrix = pyrr.matrix44.create_from_scale(pyrr.Vector3([1.0, 1.0, 1.0]))
matrix *= scale_matrix

After that, we should put this matrix into shader. In main.py this doing by pass value into

shader.set_transform(transformation())

More example you can find in pyrr official github: https://github.com/adamlwgriffiths/Pyrr

---

Load Different obj files.

Class ObjLoader from objloader folder have several function to load mesh.

To load obj file just use:

obj = ObjLoader()
vertices = obj.load_model_with_v_vt_n(path_to_obj)

Or you can combine v and vt from different obj files:

After we have created obj object when:

First function, load v (according to index) into array

obj.load_model_with_v(path_to_v)

Second function, load vt (according to index) into array.

NOTICE! In path_to_vt file there is v, vt, vn components (and indexes for every components), while in path_to_v there is just v (and index for one component v). So load vt:

obj.load_model_with_vt(path_to_vt)

Connect v and vt from different files

vertices = obj.connect_v_and_vt()

And pass this array into Mesh then we create it, like in main.py on line 69.