1 Details for Dataset Partitioning

In this section, we present the comparison of Meta-Adapter and other methods on the remaining seven datasets under different few-shot settings in Table 1. We provide the comparison of Meta-Adapter with the SOTA prompt-learning method, CoCoOp [9] in Figure 1. All experiments are conducted under the 16-shot setting. It is clear that Meta-Adapter demonstrates superior generalizability over CoCoOp by large margins. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories.