Note: If you don't see any progress (commits) don't despair, I'm still working on this (10/04/20)
This is an attempt to replicate SPACE: Unsupervised Object-Oriented Scene Representation via Spatial Attention and Decomposition https://arxiv.org/abs/2001.02407 on Pytorch. Most of the code is a clone from https://github.com/NVlabs/SSV.
A model that combines the Self-Supervised Viewpoint Learning from Image Collections (SSV) (http://arxiv.org/abs/2004.01793) encoder and the SPACE decoder is currently working on the CLEVR dataset with a very simple train look with few constraints.
Current used files are: - space-only.ipynb - utils/network_blocks.py - utils/space_modules.py - ssv.py - extern/network_blocks.py - clevr_dataset.py
For training and results please see the space-only notebook.
Took about 3 epochs to train, ~20 minutes per epoch on a GTX1080Ti. Weights used to generate this image:
https://drive.google.com/file/d/1F4KIOP2sxXrNFr33w5p1lHvo3daYWBna/view?usp=sharing