A test-time prompt-tuning method that retrieves and composes visual prompts from a learned memory bank reports 63.12% average accuracy across four ImageNet shift benchmarks, slightly ahead of prior TTA methods.
In: Proceedings of the IEEE/CVF International Conference on computer vision
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MINT: Memory-Infused Prompt Tuning at Test-time for CLIP
A test-time prompt-tuning method that retrieves and composes visual prompts from a learned memory bank reports 63.12% average accuracy across four ImageNet shift benchmarks, slightly ahead of prior TTA methods.