Given a single chair image, could we wake it up by reconstructing its 3D shape and skeleton, as well as animating its plausible articulations and motions, similar to that of human modeling?

It is a new problem that not only goes beyond image-based object reconstruction but also involves articulated animation of generic objects in 3D, which could give rise to numerous downstream augmented and virtual reality applications.

conda env create -f environment.yaml
conda activate object-wakeup
python setup.py build_ext --inplace

(Work in Progress)

Now we have a release contains sample dataset as well as the source code of the tool we developed in UNREAL 4. It is not yet finished but it is ready to play with if you are interested in.

[Data Download]

• ShapeRR_Generation contains the plug-in we developed in UE4.
• Chair_Out provides sample high-resolution rendering data for the Chairs category in ShapeNet.
• Other 3 categories presented in paper is still being processed and a full release is expected in the future.

To run a demo on our method for 3D reconstruction, please download the sample demo image from here.

Put the sample image in folder ./demo_chair.

Then you may want to download the pretrained model weights from here.

Put the checkpoint file in folder ./checkpoint.

To generate sample meshes using a trained model, use

python demo.py demo.yaml

Contact Ji Yang at [email protected] for any questions or comments.