Question about autoencoder and access to weights of trained model

[jeevster]

Hello,
Couple questions:

1. Is the convolutional autoencoder described in section 5.1 of your paper completely unsupervised, or does it incorporate bounding box label info at the bottleneck layer like the main 3D semi-supervised model?
2. Is there a flag/command line argument to only run the autoencoder and not the entire semi-supervised model?
3. Is there a way to access the weights of your trained 3D bounding box or 2D autoencoder model without running the training ourselves?

Thanks,
Sanjeev