Humans are known to benefit from geometric cues such as symmetry to group visual elements into objects for recognition. Past work in collaboration with colleagues at the University of Toronto has shown an improved ability in humans to categorize scenes from line drawings when shown contour fragments associated with more ribbon symmetric parts (versus the asymmetric parts). Although CNNs have gained much popularity recently, it is not well understood if they too can benefit from such geometric information. Using a measure of symmetry based on the medial axis, we are looking to whether CNNs share this benefit of improved category level scene classificaiton from line drawings, when symmetry is emphasized.