Embodied Language Grounding With 3D Visual Feature Representations

IMPORTANT: This is our updated paper and repository, Please use this instead:

https://github.com/mihirp1998/Disentangling-3D-Prototypical-Nets

As this code only works for 64x64 images, where as the new code works on 256x256 images and also doesn't require as many labels.

Project page

https://mihirp1998.github.io/project_pages/emblang/

Paper

https://arxiv.org/abs/1910.01210

Code

cd vis_imagine_static_voxels

python main.py gqn3d_full_model_dual_loss_3dmask_l2