UniFa: A unified feature hallucination framework for any-shot object detection

Hui Nie, Ruiping Wang, Xilin Chen

Abstract: Highlights•Employ generative techniques to address the any-shot object detection task.•Enhance visual features via visual-aware semantic information.•Design a feature enhancer to construct a more realistic feature distribution.•Reduce skewed predictions using a zero-shot class discriminator.•Achieves the new SoTA performance in any-shot object detection.