We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: A Comparative Review of Recent Few-Shot Object Detection Algorithms

Abstract: Few-shot object detection, learning to adapt to the novel classes with a few labeled data, is an imperative and long-lasting problem due to the inherent long-tail distribution of real-world data and the urgent demands to cut costs of data collection and annotation. Recently, some studies have explored how to use implicit cues in extra datasets without target-domain supervision to help few-shot detectors refine robust task notions. This survey provides a comprehensive overview from current classic and latest achievements for few-shot object detection to future research expectations from manifold perspectives. In particular, we first propose a data-based taxonomy of the training data and the form of corresponding supervision which are accessed during the training stage. Following this taxonomy, we present a significant review of the formal definition, main challenges, benchmark datasets, evaluation metrics, and learning strategies. In addition, we present a detailed investigation of how to interplay the object detection methods to develop this issue systematically. Finally, we conclude with the current status of few-shot object detection, along with potential research directions for this field.
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as: arXiv:2111.00201 [cs.CV]
  (or arXiv:2111.00201v1 [cs.CV] for this version)

Submission history

From: Jiaxu Leng [view email]
[v1] Sat, 30 Oct 2021 07:57:11 GMT (20381kb,D)

Link back to: arXiv, form interface, contact.