Object identity determines transsaccadic integration

Very little information is transferred across saccades. It is commonly thought that detailed vision starts mainly anew with each saccade. Here, we show that transsaccadic integration occurs even for very fine grained and unconscious information when object identity prevails across saccades through non-retinotopic reference-frames. We presented a single line followed by subsequent lines, one each of its sides creating the impression of two expanding motion streams. When one line is offset in one of the streams (Vernier offset), all lines in this, and only this stream, appear to be offset. When a second line in this stream is offset into the opposite direction, the two offsets cancel each other-even when the offsets are presented 400ms apart from each other. Integration is mandatory and unconscious, i.e., observers cannot tell which lines and how many lines are offset. Surprisingly, mandatory integration occurs also when observers make a saccade during the stream motion-even when one offset is presented before and the other one after the saccade. Importantly, the before-and-after-the-saccades parts of the streams are presented at very different locations on the retina. When these two parts of the streams are presented at the same retinal locations but observers make no saccade, two parts of two streams with two independent offsets are perceived. There is no integration of the motion streams and offsets. Hence, the human brain first establishes a motion reference-frame that unites the two parts of the motion streams to preserve object identity and then it integrates the offsets across space, time, and retinal locations according to prevailing object identities. Our results cannot easily be explained by classic remapping because integration depends on the trajectory of the streams, which, at the time of the saccade is not granted.