Skip to content

Latest commit

 

History

History
88 lines (62 loc) · 2.4 KB

README.md

File metadata and controls

88 lines (62 loc) · 2.4 KB

rlviewer

A fast 3D renderer for the reinforcement learning

rlviewer is a Python module that loads a .OBJ file, renders it using the OpenGL API, and returns the resulting view as a numpy array. The code is derived from the at OpenGL tutorials (code).

Installation

The module requires the following dependencies (most will already be installed):

sudo apt install  cmake make g++ libx11-dev libxi-dev libgl1-mesa-dev libglu1-mesa-dev libxrandr-dev libxext-dev libxcursor-dev libxinerama-dev libxi-dev libglfw3-dev libglfw3-dev libglew2.0 libglew-dev

To build the code:

python3 setup.py build

Install the module:

System-wide:

sudo python3 setup.py install

Or for the current user:

python3 setup.py install --user

Usage

Once the module is imported (import rlviewer), there are two functions available: rlviewer.load(path) and rlviewer.grab(radius, latitude, longitude).

The load function imports a .OBJ file. The 3D object should be centered on the origin. The obj loader is very basic and you may run into issues. (TODO: fix issues :) and/or document work arounds...)

The grab function renders the scene. The camera moves on a sphere around the object. You must specify the radius of the sphere and the camera's position on the sphere with the latitude (vertical angle) and longitude (horizontal angle).

The function returns a numpy array that represents the rendered scene. It's a RGB images and its size is currently fixed at 1024 x 768 pixels. So you get a np.array((1024, 768, 3), dtype=np.uint8) matrix in return.

In the file render.py you find a complete example:

import rlviewer
import cv2
import numpy as np
import os
 
output_dir = "out"
if not os.path.exists(output_dir):
    os.makedirs(output_dir)
  
rlviewer.load('215.obj') 

for lat in range(0, 90, 5): 
    for lon in range(0, 360, 3):
        # Take a picture of the object positioned in the origin with
        # the camera moving on a sphere with a radius of 20 and
        # positioned at the given latitude and longitude.
        img = rlviewer.grab(20.0, np.radians(lat), np.radians(lon)) 
        cv2.imwrite(f"{output_dir}/image-{lat:03d}-{lon:03d}.png", img)

Acknowledgements

Thank you to the OpenGL tutorials! The original OpenGL source is available under the WTF license.