cavanagh's pseudorealism Jan 23 - David Thompson
(Outline) 1. Marr's Model 2. Cavanagh's Model 3. What more can artists tell us about vision?
the Marr vision model
input image
the Marr vision model primal sketch
the Marr vision model 2.5-D sketch
the Marr vision model 3D model
stimulus 2½D sketch Object 3D model primal sketch viewer-centric coordinates object-centric coordinates the Marr vision model
stimulus 2½D sketch Object 3D model bottom-up reconstruction of shape from image features recognition with object- centric 3D models primal sketch viewer-centric coordinates object-centric coordinates the Marr vision model
Geons (Biederman '87)
Cavanagh's model
two-tone images
hair (not shadow!) inferred external contours “attached shadow” contour “cast shadow” contour
A. Finding 3D structure in two-tone images requires distinguishing cast shadows, attached shadows, and areas of low reflectivity B. The images do not contain this information a priori D. Therefore, our recognition process must involve matching viewpoint-specific 2D templates Cavanagh's claim
What else can art tell us about object recognition? Cavanagh's model
What else can art tell us about object recognition? Cavanagh's model neurobiologic al evidence! (Logothetis '95)
What else can art tell us about object recognition? Cavanagh's model stimulus 2D shape memory 3D shape top-down feedback basic recognition with 2D primitives
shadows and lighting
transparen cy
humans don't do ray tracing
line drawings are universal
recognition with few cues
Human object recognition uses shortcuts, not complete physical models - local lighting - simplified reflections At least some of our processing happens through “simple vision” - viewer-centric 2D template matching - simple 3D memory provides top-down feedback take-home lessons?
Marr's model (circa 1980) Cavanagh's model (circa 1990) Two models of Human visual processing stimulu s 2D shape memor y 3D shape 2½D sketch Objec t 3D model feedback basic recognition with 2D primitives reconstruction of shape from image features object recognition by matching 3D models primal sketch