Model performance is sensitive to disruptions in alignment beyond a simple heuristic of orientation similarity and proximity.
<p><b>(A)</b> Schematic showing manipulation of alignment (‘good continuation’) while maintaining orientation similarity and local proximity between contour elements. This is done by displacing each contour element with an alignment jitter in a direction perpendicular to the local...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | , |
| Published: |
2025
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p><b>(A)</b> Schematic showing manipulation of alignment (‘good continuation’) while maintaining orientation similarity and local proximity between contour elements. This is done by displacing each contour element with an alignment jitter in a direction perpendicular to the local contour path. <b>(B</b>) Example stimuli showing a contour with 0 pixels of jitter (left) and 8 pixels of jitter (right). With alignment hitter of 8 pixels, it is much harder to detect the contour in the display. <b>(C)</b> A visualization of the isolated contour elements for each alignment jitter value (0, 2, 4, 6, and 8 pixels). <b>(D</b>) Performance of broadly-tuned model on the jittered stimuli. <b>(E)</b> Performance of narrowly-tuned model (fine-tuned on 20° contours) on the jittered stimuli. Performance systematically declines for both fine-tuned models as alignment jitter increases.</p> <p>(TIF)</p> |
|---|