Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Few-Shot Domain Adaptive Object Detection for Microscopic Images

Sumayya Inayat, Nimra Dilawar, Waqas Sultani, Mohsen Ali

TL;DR

The paper tackles few-shot domain adaptive object detection for microscopic images, addressing domain shift and extreme class imbalance that hamper generalization. It introduces Class Balancing Cut-Paste (CBCP) to rebalance data and Intra-Inter Domain Feature Alignment (I2DA) to align instance features across and within domains, augmented by an instance-level classifier loss. The approach yields state-of-the-art results on M5 Malaria and Raabin-WBC under 2-, 3-, and 5-shot settings, with notable gains in mAP@50 and recall and robust qualitative improvements. Released code further facilitates adoption in biomedical imaging with limited labeled data and domain variation.

Abstract

In recent years, numerous domain adaptive strategies have been proposed to help deep learning models overcome the challenges posed by domain shift. However, even unsupervised domain adaptive strategies still require a large amount of target data. Medical imaging datasets are often characterized by class imbalance and scarcity of labeled and unlabeled data. Few-shot domain adaptive object detection (FSDAOD) addresses the challenge of adapting object detectors to target domains with limited labeled data. Existing works struggle with randomly selected target domain images that may not accurately represent the real population, resulting in overfitting to small validation sets and poor generalization to larger test sets. Medical datasets exhibit high class imbalance and background similarity, leading to increased false positives and lower mean Average Precision (map) in target domains. To overcome these challenges, we propose a novel FSDAOD strategy for microscopic imaging. Our contributions include a domain adaptive class balancing strategy for few-shot scenarios, multi-layer instance-level inter and intra-domain alignment to enhance similarity between class instances regardless of domain, and an instance-level classification loss applied in the middle layers of the object detector to enforce feature retention necessary for correct classification across domains. Extensive experimental results with competitive baselines demonstrate the effectiveness of our approach, achieving state-of-the-art results on two public microscopic datasets. Code available at https://github.co/intelligentMachinesLab/few-shot-domain-adaptive-microscopy

Few-Shot Domain Adaptive Object Detection for Microscopic Images

TL;DR

The paper tackles few-shot domain adaptive object detection for microscopic images, addressing domain shift and extreme class imbalance that hamper generalization. It introduces Class Balancing Cut-Paste (CBCP) to rebalance data and Intra-Inter Domain Feature Alignment (I2DA) to align instance features across and within domains, augmented by an instance-level classifier loss. The approach yields state-of-the-art results on M5 Malaria and Raabin-WBC under 2-, 3-, and 5-shot settings, with notable gains in mAP@50 and recall and robust qualitative improvements. Released code further facilitates adoption in biomedical imaging with limited labeled data and domain variation.

Abstract

In recent years, numerous domain adaptive strategies have been proposed to help deep learning models overcome the challenges posed by domain shift. However, even unsupervised domain adaptive strategies still require a large amount of target data. Medical imaging datasets are often characterized by class imbalance and scarcity of labeled and unlabeled data. Few-shot domain adaptive object detection (FSDAOD) addresses the challenge of adapting object detectors to target domains with limited labeled data. Existing works struggle with randomly selected target domain images that may not accurately represent the real population, resulting in overfitting to small validation sets and poor generalization to larger test sets. Medical datasets exhibit high class imbalance and background similarity, leading to increased false positives and lower mean Average Precision (map) in target domains. To overcome these challenges, we propose a novel FSDAOD strategy for microscopic imaging. Our contributions include a domain adaptive class balancing strategy for few-shot scenarios, multi-layer instance-level inter and intra-domain alignment to enhance similarity between class instances regardless of domain, and an instance-level classification loss applied in the middle layers of the object detector to enforce feature retention necessary for correct classification across domains. Extensive experimental results with competitive baselines demonstrate the effectiveness of our approach, achieving state-of-the-art results on two public microscopic datasets. Code available at https://github.co/intelligentMachinesLab/few-shot-domain-adaptive-microscopy
Paper Structure (7 sections, 4 equations, 3 figures, 3 tables)

This paper contains 7 sections, 4 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Class Balancing Domain Generalized Cut-Paste
  4. Inter-domain Alignment and Intra-domain Class Consistency
  5. Experiments and Results
  6. Results
  7. Conclusion

Figures (3)

  • Figure 1: Class Balancing Cut Paste strategy: We first compute the metadata and increment-stats from the abundant source dataset and few target images and then increment cells to the images with less pre-existing cells.
  • Figure 2: Proposed approach: We first build our class-wise balanced dataset through a cut-paste strategy (Fig \ref{['fig:cut-paste-flow']}), then train the model with our proposed inter-domain instance feature-level alignment and intra-domain instance feature-level consistency. We extract multi-layer neck features and upsample them to a common size, followed by the extraction of pooled object-level features, which are then passed to the similarity-dissimilarity and classification module.
  • Figure 3: Qualitative results of LCM Malarial defected regions after adaptation.