Vision Models Predict Urban Scene Appraisal with Limited Neural Alignment

Name: Vision Models Predict Urban Scene Appraisal with Limited Neural Alignment
Uploaded: 2026-10-09T03:49:00.976597+00:00
Duration: 511 s
Description: Researchers tested if pretrained vision embeddings organize urban scenes like human perception does by comparing them to brain data geometry. While DINOv2 ViT-B is the best model, there is no reliable link between how well a model predicts human ratings and how closely its internal structure matches the neural geometry.

Researchers tested if pretrained vision embeddings organize urban scenes like human perception does by comparing them to brain data geometry. While DINOv2 ViT-B is the best model, there is no reliable link between how well a model predicts human ratings and how closely its internal structure matches the neural geometry.

Open the full episode