Melissa Allouche¹ (), Lisa Schwetlick^1,2, Michael Herzog¹; ¹Laboratory of Psychophysics, Brain Mind Institute, École Polytechnique Fédérale de Lausanne (EPFL), ²École Normale Supérieure

In crowding, perception of a stimulus deteriorates when flankers are added around it. Feedforward models of vision predict that as more flankers are added, crowding increases and thus performance further decreases. However, increasing the number or complexity of flankers can improve performance, a phenomenon known as uncrowding. This occurs especially when flankers form coherent structures (e.g., a cube versus the scrambled lines of the cube). Uncrowding requires stimuli to be presented for a minimum of 160ms before it emerges. It has been proposed that this comparatively long time suggests the involvement of feedback processes. Using backward masking, a tool that has been shown to disrupt feedback, we investigate whether uncrowding can be disrupted. Participants performed a vernier discrimination task in three conditions: vernier in isolation (baseline), vernier flanked by lines (typically crowded), and vernier flanked by cubes (typically uncrowded) where the target was presented for 120ms. The vernier was followed by a mask of randomly oriented lines. The mask reduces performance in all conditions, but this effect is particularly strong in the cubes condition, where uncrowding is clearly disrupted. A mask duration of as little as 10ms is sufficient to induce this effect. If a flanker-only preview is presented, i.e., the cubes without the vernier, ideally 60-120ms before the target, participants perform as well as in the no mask condition. Our first result demonstrates that after a presentation time of 120ms, the cube perception, and therefore uncrowding, is still fragile and can be disrupted by a mask. Our second result shows that this processing can be kick-started with 20ms flanker-only previews and suggests that once this processing is complete, the cubes perception becomes robust and immune to the mask. Together, the preview and the mask provide a toolkit to investigate the time course of object recognition.