Perceptual misalignment of texture representations in convolutional neural networks

Name: Perceptual misalignment of texture representations in convolutional neural networks
Uploaded: 2026-10-10T15:19:46.837034+00:00
Duration: 887 s
Description: The study tested how well feature correlations from Convolutional Neural Networks (CNNs) capture human texture perception. It found no link between conventional CNN quality metrics and how well these representations align with human texture judgment, suggesting standard CNN measures do not reflect perceptual alignment.

The study tested how well feature correlations from Convolutional Neural Networks (CNNs) capture human texture perception. It found no link between conventional CNN quality metrics and how well these representations align with human texture judgment, suggesting standard CNN measures do not reflect perceptual alignment.

Open the full episode