Predicting and verifying regions of interest selections

A method for determining regions of interest (ROIs) in video images, comprising:

• generating user viewing behavior data while rendering a video clip including a video image in a plurality of viewports of a plurality of content viewers, wherein each content viewer in the plurality of content viewers views the video clip including the video image through a respective viewport in the plurality of viewports;
• determining, from the user viewing behavior data, a plurality of spatial locations as candidate ROIs, in the video image and other video images of the video clip, to which a plurality of foveal visions of the plurality of content viewers is directed;
• identifying, based on the plurality of spatial locations in the video image and the other images of the video clip, one or more unique spatial paths represented in the video clip, wherein the one or more unique spatial paths are identified by applying one or more computing techniques to the plurality of spatial locations in the video image and the other images of the video clip;
• using the one or more unique spatial paths represented in the video clip to identify at least one of two or more ROIs in the video image;
• encoding two or more image portions in the two or more ROIs of the video image and one or more other image portions outside the two or more ROIs of the video image into a video stream;
• wherein the two or more image portions are non-overlapping with the one or more other image portions;
• wherein the two or more image portions in the two or more ROIs of the video image are encoded in the video stream at spatiotemporal resolutions higher than other spatiotemporal resolutions at which the one or more other image portions outside the two or more ROIs of the video image are encoded in the video stream;
• transmitting, to a streaming client device, the video stream.