A familiar-size Stroop effect in the absence of basic-level recognition

When we view a picture of an object, we automatically recognize what the object is and know how big it typically is in the world (Konkle & Oliva, 2012). Is information about an object's size activated only after we've identified the object, or can this size information be activated before object recognition even occurs? We previously found that big and small objects differ in mid-level perceptual features (Long, Konkle, Cohen, & Alvarez, 2016). Here we asked whether these perceptual features can automatically trigger real-world size processing, bypassing the need for basic-level object recognition. To test this hypothesis, we used an image synthesis algorithm to generate "texform" images, which are unrecognizable versions of big and small objects that still preserve some textural and form information from the original images. Across two experiments, we find that even though these synthesized stimuli cannot be identified, they automatically trigger familiar size processing and give rise to a Size-Stroop effect. Furthermore, we isolate perceived curvature as one feature the visual system uses to infer real-world size. These results suggest that mid-level perceptual features can automatically feed forward to facilitate object processing, and challenge the idea that we must first identify an object before we can access its higher-level properties.