Skip to content

Latest commit

 

History

History

ycb_render

README.md

YCB Renderer

This is an PyOpenGL version of YCB renderer, it has some key features to work with deep learning frameworks. This renderer is created by Fei Xia ([email protected]) and modified by Lirui Wang ([email protected]) and Yu Xiang ([email protected]).

  • EGL headless rendering, so a running x server is not required
  • Rendering to multiple texture buffers, so it can render rgb, segmentation, point cloud in a single shader call
  • Uses OpenGL - CUDA interoperation to render to pytorch tensor directly and can render 1024x1024 images at 1300 fps.
  • Support eglQueryDevicesEXT and eglGetPlatformDisplayEXT via pybind11. So it is easy to choose which GPU to use.
  • Minimalistic, it uses raw OpenGL, doesn't rely on third-party GL helper libraries other than assimp.
  • It supports rendering of textured structured objects and simple geometry such as points/lines as well as other features.
  • It uses cv2 as a minimal interactive app for mouse/keyboard control.
  • Parallelism is used to speed up loading.

Installation

Install dependencies:

sudo apt-get install libassimp-dev

In a virtual environment, run python setup.py develop should compile c++ sources and setup paths.

Example

Run ./download_data.sh in the root directory of the OMG package to download necessary data.

python ycb_renderer.py ycb for rendering YCB objects

python ycb_renderer.py panda_arm for rendering Franka Panda arm model

python ycb_renderer.py shapenet for rendering an example shapenet mug

or aribitray path to the model directory or .obj file or lists of objects (split by ,) of you like with python ycb_renderer.py ../data/objects/002_master_chef_can/,../data/objects/003_cracker_box/. Mouse window interaction:

(control + left) to rotate
(control + middle) to translate 
(shift + left) to zoom

Keyboard window interaction:

'esc' to quit
's' to interact
'n' to acquire camera pose
'x' to reset camera
'q' to tilt up
'e' to tilt down
'a' to pan left
'd' to pan right
'z' to zoom out 
'c' to zoom in 
'w' to toggle background color
'p' to take snapshot
'v' to toggle video recording
'i,k,j,l,m,<,r' to change object pose
'tab' to switch color, segmentation, normal, and points 
'`' to toggle wireframe and point
'1' to toggle vertex normal drawing
'2' to capture observed points
'3' to aggregate capture observed points
'space' to render coordinate axis
python ycb_renderer.py panda_arm python ycb_renderer.py shapenet2
python ycb_renderer.py ycb python ycb_renderer.py shapenet