Appendix for " Learning Dynamic Attribute-Factored World Models for Efficient Multi-object Reinforcement Learning "