Steerable Visual Representations
Paper • 2604.02327 • Published • 55
Steerable Visual Representations (SteerViT)
SteerViT is a framework that allows you to guide a pretrained vision model's focus and features using natural language
-
JonaRuthardt/SteerViT
Image Feature Extraction • Updated • 7Note Model Checkpoints
JonaRuthardt/SteerViT
Viewer • Updated • 2.29M • 202 • 2Note Training Dataset
JonaRuthardt/CORE
Viewer • Updated • 2.39k • 205Note CORE Benchmark to measure steerability
SteerViT
👁10Steering Vision Representations with Language
Note Interactive Demo