MAIR++: Improving Multi-view Attention Inverse Rendering with Implicit Lighting Representation | Synapse