Few-Shot Domain Adaptive Object Detection for Microscopic Images

(Published at MICCAI 2024)

Sumayya Inayat,    Nimra Dilawar,  Waqas Sultani,    Mohsen Ali
Intelligent Machines Lab, Information Technology University, Pakistan

Paper

Github Logo-03

Dataset/Code

poster

Poster

slides

Slides

Motivation
Few-shot domain adaptive object detection (FSDAOD) addresses the challenge of adapting object detectors to target domains with limited labeled data. We propose a novel FSDAOD strategy for microscopic imaging. Our contributions include: 1) a domain adaptive class balancing strategy for few shot scenarios; 2) multi-layer instance-level inter and intra-domain alignment by enhancing similarity between the instances of classes regardless of the domain and enhancing dissimilarity when it’s not. Furthermore, an instance-level classification loss is applied in the middle layers of the object detector to enforce the retention of features necessary for the correct classification regardless of the domain. Extensive experimental results with competitive baselines indicate the effectiveness of our proposed framework by achieving state-of-the-art results on two public microscopic datasets.
Proposed approach: We first build our class-wise balanced dataset through a cut-paste strategy, 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.
CBCP effectively handles class imbalance by strategically augmenting source and target cells based on target-domain visuals, leading to these improved results. 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.
Results
Qualitative Results
About Authors

Sumayya Inayat

Research Associate, Intelligent Machines Lab, ITU, Lahore, Pakistan

Email: sumayya.inayat@itu.edu.pk
LinkedIn: Sumayya Inayat

Nimra Dilawar

Ph.D. Fellow, Intelligent Machines Lab, ITU, Lahore, Pakistan
 
LinkedIn: Nimra Dilawar

Waqas Sultani

Assistant Professor, Department of AI, ITU, Lahore, Pakistan
 
LinkedIn: Waqas Sultani

Mohsen Ali

Associate Professor Tenured, Department of AI, ITU,  Pakistan
 
LinkedIn: Mohsen Ali

BibTex

@article{inayat2024few,
  title={Few-Shot Domain Adaptive Object Detection for Microscopic Images},
  author={Inayat, Sumayya and Dilawar, Nimra and Sultani, Waqas and Ali, Mohsen},
  journal={arXiv preprint arXiv:2407.07633},
  year={2024}
}