IC-934DNN object detectors do not exhibit the same object size effect as humans in peripheral object detection

Anne Harrington, Vasha DuTell, Mark Hamilton, Ayush Tewari, Simon Stent, William T. Freeman, Ruth Rosenholtz

SourceCOCO-Periph: Bridging the Gap Between Human and Machine Perception in the Periphery

Human critical eccentricity for peripheral object detection increases with the size of the target object, meaning humans detect larger objects at greater eccentricities. This relationship does not appear for any of the tested DNN object detection models, whose critical eccentricities remain low for both small and large objects. The absence of a size effect persists even in models trained on COCO-Periph that show higher AP-large and AP-small than baseline on TTM-transformed images.

Evidence
correlational
Caveat
The relationship is assessed qualitatively from Figure 6a; no specific correlation values or effect sizes are reported in the text for the DNN size effect.
Model
DINO-FocalNet-Large, Swin Transformer DINO-Swin-Tiny, DETR-R50, RetinaNet-R50, FoveaBox, Faster R-CNN / Faster R-CNN R50 / Faster R-CNN X101
Datasets
MS COCO / COCO / COCO 2014 / COCO 2017 / COCO 20k / COCO-it / COCO-wl [source]
Methods
Uniform TTM [primary]
Related findings
IC-932, IC-933
Extraction
automatic-extraction