FSP-DETR: Few-Shot Prototypical Parasitic Ova Detection

2025-10-10

AI Summary

Biomedical object detection suffers from scarce annotated data and frequent emergence of novel or rare categories. To address this, we propose FSP-DETR—a unified framework that jointly integrates few-shot detection, open-set recognition, and cross-task generalization within a single model, enabling zero-shot inference on unseen classes and background rejection without fine-tuning. Our approach builds upon a class-agnostic DETR backbone; it generates class prototypes from support images, enhances representation robustness via augmented views, a lightweight Transformer decoder, and KL-divergence regularization; and jointly optimizes prototype matching and separation losses to enable dynamic task-mode switching at inference time. Evaluated on parasite egg, blood cell, and malaria detection tasks, FSP-DETR significantly outperforms state-of-the-art few-shot and prototypical methods—especially under low-shot (1–5 shots) and open-set settings. We further release a new benchmark dataset comprising 20 parasite egg classes.

Technology Category

• Computer Vision: Object Detection & Categorization
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Security and Privacy: Data transparency and provenance

Abstract

Object detection in biomedical settings is fundamentally constrained by the scarcity of labeled data and the frequent emergence of novel or rare categories. We present FSP-DETR, a unified detection framework that enables robust few-shot detection, open-set recognition, and generalization to unseen biomedical tasks within a single model. Built upon a class-agnostic DETR backbone, our approach constructs class prototypes from original support images and learns an embedding space using augmented views and a lightweight transformer decoder. Training jointly optimizes a prototype matching loss, an alignment-based separation loss, and a KL divergence regularization to improve discriminative feature learning and calibration under scarce supervision. Unlike prior work that tackles these tasks in isolation, FSP-DETR enables inference-time flexibility to support unseen class recognition, background rejection, and cross-task adaptation without retraining. We also introduce a new ova species detection benchmark with 20 parasite classes and establish standardized evaluation protocols. Extensive experiments across ova, blood cell, and malaria detection tasks demonstrate that FSP-DETR significantly outperforms prior few-shot and prototype-based detectors, especially in low-shot and open-set scenarios.

Problem

• Detects parasitic ova with limited labeled data
• Enables open-set recognition for unseen biomedical categories
• Generalizes across tasks without retraining through prototype learning

Innovation

• Unified DETR framework for few-shot biomedical detection
• Class prototypes from support images with transformer decoder
• Joint optimization with matching, separation, and regularization losses