vc-shapenet

ShapeNet rendering scripts for visual concept

Install

Download ShapeNetCore.v1.zip to this folder.

Unzip the zip file, the folder structure will looks like ShapeNetCore.v1/02691156/[model_id]. model_id is a very long string and is the unique id for a model.
Setup code from RenderForCNN
- git submodule init and git submodule update to get RenderForCNN code into the rendercnn folder.
- mv global_variables.py.example to global_variables.py and change the configuration for blender
Notice: The dependency on RenderForCNN is quite heavy and not necessary at all. we can consider get rid of it.

render_airplane.py

Batch script to render all airplane model, use the viewpoint file in viewpoints folder
view_3d_model.py

Take [model_id] as its input argument. Show this 3D model in blender.
show_preview.py

Generate an html file containing rendered images.

blender_script.py

The code to be executed in blender, responsible for setting up camera, loading model, etc.
render_opt.py

Rendering configuration code, such as lighting, etc.

The viewpoint is saved in viewpoints/ folder.

In addition to Weichao's code, render all categories of more view angles from ShapeNet.hhhh

In addition to Weichao's, I recommend adding GNU-parallel module to speed up rendering process(OPTIONAL).

sudo apt-get parallel

render_all.py

Batch script to render all all models from all categories, use the all_views.txt file in viewpoints folder
blender_all.py

Fixing the lighting problem in previous blender script
name.txt

Listing the name of categories

python render_all.py

After a few minites, you will see render_commands.txt in the root folder.

parallel < render_commands.txt

This takes about 8 days to run in 12 cores in parallel. If you don't want it in parallel, run render commands as shell scripts.

##Output

Images will be populated in ROOT_FOLDER/images/. The filename convention is [category_id]_[model_id]_a[azimuth]_e[elevation]_t[tilt]_d[distance].png.