It has often been remarked upon that the eyes in frontal portraits of people "follow the observer around the room". Although there has been much speculation in both the vision and art history communities, there exists comparatively little quantitative data on the topic. We have prepared photographs of the depiction of a human torso framed within a baroque (3D relief) gilded frame, hung on a flat brick wall. Realistic illumination lets the frame appear solid, detached from the wall, containing a flat picture. These photographs were presented on a flat screen monitor at the correct apparent size, either undeformed, or projectively (pre-)deformed, with the screen being viewed monocularly, either frontally or obliquely. We set things up such that the retinal image can be deformed or undeformed in either case. We have measured the apparent spatial orientation (slant and tilt) of both the wall and many surface elements of the pictorial relief of the torso. Such measurements allow us to study the perception of both the torso and the wall quantitatively. In an additional experiment we determined the apparently frontoparallel points on the pictorial relief, this immediately relates to the effect of protrusions "following the observer around the room". We find that observers do indeed correct for the effects of oblique viewing, be it only to a certain extent. There exist marked individual differences. The "corrections" applied by the observers are composed of stretches and shears of pictorial space, thus exploiting the ambiguity left by the monocular depth cues.