好的，请发送需要翻译的内容。

### Direct Translation In recent years, with the rapid development of deep learning technology, zero-shot and few-shot learning have become research hotspots in the field of computer vision. These methods aim to enable models to recognize new categories without or with only a small amount of labeled data. However, existing methods still face many challenges when dealing with complex scenes, such as domain shift and semantic gap issues. To address these problems, we propose a new framework that effectively combines the advantages of zero-shot and few-shot learning, significantly improving the model's generalization ability in unseen categories. *** ### Problems 1. "research hotspots" is slightly informal for a scientific paper; "research focuses" or "research frontiers" would be more academic. 2. "without or with only a small amount of labeled data" is awkward; the phrasing is not smooth and could be reorganized for clarity. 3. "domain shift and semantic gap issues" is acceptable but could be more precise; "the issues of domain shift and semantic gap" reads better. 4. "we propose a new framework" is fine, but "effectively combines" could be more formal, e.g., "integrates" or "synergizes." 5. "significantly improving the model's generalization ability" is clear but could be more concise and idiomatic, e.g., "substantially enhancing generalization performance." *** ### Paraphrase ``` In recent years, with the rapid advancement of deep learning, zero-shot and few-shot learning have emerged as prominent research focuses in computer vision. These paradigms aim to enable models to recognize novel categories with little or no labeled data. Nevertheless, existing approaches still encounter significant challenges in complex scenarios, including domain shift and semantic gap. To address these limitations, we introduce a novel framework that integrates the strengths of both zero-shot and few-shot learning, thereby substantially enhancing the model's generalization capability on unseen categories. ```