Connectome datasets -- A connectome (/kəˈnɛktoʊm/) is a comprehensive map of neural connections in the brain. (via Wikipedia). Can contain functional connectivity and anatomical connectivity.
Deep Image Reconstruction: This is the study where authors used Artificial Neural Network (Deep Learning) to reconstruct the images that the subject sees.
Generic Object Decoding (fMRI on ImageNet): visual features can be predicted from fMRI patterns, which in turn can be used to predict mental images (which can be an imaginary image or something the subject is looking at). Reference.
Non-image data
There are other non-image data sources that might be of interest: