Multi-View Part-Based Few-Shot Object Detection.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Few-shot object detection (FSOD) aims to identify new object categories using limited annotated data.
• A significant challenge in FSOD is object misclassification due to insufficient discriminative information learned from few samples.

Purpose of the Study:

• To address the object misclassification issue in FSOD.
• To propose a novel multi-view part-based FSOD network (MPFSOD) for more accurate detection of new object categories.

Main Methods:

• Developed a part-based detector (PBD) to generate task-aware object-level parts for enhanced representation.
• Introduced an image-level multi-view fusion module (Img-MVF) and an instance-level multi-view modulation module (Inst-MVM) to extract richer discriminative information from multiple views.

Main Results:

• The proposed MPFSOD method significantly improves performance on PASCAL VOC and MS COCO datasets.
• Achieved up to 11.2% improvement over a strong baseline and 4.3% average improvement over state-of-the-art methods.

Conclusions:

• The MPFSOD network effectively generates highly discriminative parts by leveraging multi-view information.
• The approach enhances object characterization, leading to more accurate few-shot object detection and reduced misclassification.