Concept Grounding with Modular Action-Capsules in Semantic Video Prediction

Preprint. Under review.

Authors Anonymous.  
[Paper]
[Code](coming soon)
[Bibtex]

Video Overview




Our Solution: Modular Action Capsule Network (MAC)






Qualitative Results on TowerCreation

 Groundtruth      Best sample       Sample 1         Sample 2  

 Groundtruth      Best sample       Sample 1         Sample 2  




Qualitative Results on CLEVR-Building-blocks

Ground truth                    Predictions

Ground truth                    Predictions




Qualitative Results on Sapien-Kitchen

Ground truth                    Predictions

Ground truth                    Predictions




Quantitative Comparison






Counterfactual Generation with displayed labels






Acknowledgements

To be added after review period.