Is Less More? Exploring Token Condensation as Training-free Test-time Adaptation

VLM
Re-assign community