Category Knowledge-Guided Parameter Calibration for Few-Shot Object Detection.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Few-shot object detection (FSOD) addresses the challenge of adapting generic object detectors to new categories with minimal annotated data.
• While generic object detection is well-researched, FSOD remains an under-explored but crucial area for realistic applications.
• Existing methods often struggle with the limited data availability inherent in FSOD tasks.

Purpose of the Study:

• To propose a novel Category Knowledge-guided Parameter Calibration (CKPC) framework to address the challenges in few-shot object detection.
• To enhance the performance of generic object detectors on novel categories using limited annotations.
• To improve the adaptability and robustness of object detection models in low-data regimes.

Main Methods:

• Propagating category relation information to derive representative category knowledge.
• Exploiting RoI-RoI and RoI-Category relations to capture local-global context and enhance Region of Interest (RoI) features.
• Projecting foreground category knowledge into a parameter space for classifier generation and learning a proxy background category for improved discrimination.

Main Results:

• The proposed CKPC framework successfully calibrates instance-level classifiers using generated category-level parameters.
• Extensive experiments on Pascal VOC and MS COCO benchmarks demonstrate superior performance compared to state-of-the-art methods.
• The framework effectively improves detection performance in few-shot scenarios.

Conclusions:

• The CKPC framework offers a significant advancement in few-shot object detection by effectively leveraging category knowledge.
• The proposed approach provides a robust solution for adapting object detectors to novel categories with limited data.
• This work paves the way for more efficient and accurate object detection in data-scarce environments.