Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detect an Object At Once without Fine-tuning

Junyu Hao, Jianheng Liu, Yongjia Zhao, Zuofan Chen, Qi Sun, Jinlong Chen, Jianguo Wei, Minghao Yang

TL;DR

By pre-learning from labels annotated in traditional datasets, the SDM-RAN can detect previously unknown objects without fine-tuning and outperforms state-of-the-art methods on the same task.

Abstract

When presented with one or a few photos of a previously unseen object, humans can instantly recognize it in different scenes. Although the human brain mechanism behind this phenomenon is still not fully understood, this work introduces a novel technical realization of this task. It consists of two phases: (1) generating a Similarity Density Map (SDM) by convolving the scene image with the given object image patch(es) so that the highlight areas in the SDM indicate the possible locations; (2) obtaining the object occupied areas in the scene through a Region Alignment Network (RAN). The RAN is constructed on a backbone of Deep Siamese Network (DSN), and different from the traditional DSNs, it aims to obtain the object accurate regions by regressing the location and area differences between the ground truths and the predicted ones indicated by the highlight areas in SDM. By pre-learning from labels annotated in traditional datasets, the SDM-RAN can detect previously unknown objects without fine-tuning. Experiments were conducted on the MS COCO, PASCAL VOC datasets. The results indicate that the proposed method outperforms state-of-the-art methods on the same task.

Detect an Object At Once without Fine-tuning

TL;DR

By pre-learning from labels annotated in traditional datasets, the SDM-RAN can detect previously unknown objects without fine-tuning and outperforms state-of-the-art methods on the same task.

Abstract

When presented with one or a few photos of a previously unseen object, humans can instantly recognize it in different scenes. Although the human brain mechanism behind this phenomenon is still not fully understood, this work introduces a novel technical realization of this task. It consists of two phases: (1) generating a Similarity Density Map (SDM) by convolving the scene image with the given object image patch(es) so that the highlight areas in the SDM indicate the possible locations; (2) obtaining the object occupied areas in the scene through a Region Alignment Network (RAN). The RAN is constructed on a backbone of Deep Siamese Network (DSN), and different from the traditional DSNs, it aims to obtain the object accurate regions by regressing the location and area differences between the ground truths and the predicted ones indicated by the highlight areas in SDM. By pre-learning from labels annotated in traditional datasets, the SDM-RAN can detect previously unknown objects without fine-tuning. Experiments were conducted on the MS COCO, PASCAL VOC datasets. The results indicate that the proposed method outperforms state-of-the-art methods on the same task.

Paper Structure

This paper contains 26 sections, 8 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Few-shot Object Detection
  4. Few-shot Object Detection At Once (FSOD-AO)
  5. Methodology
  6. Framework
  7. Definition
  8. Workflow
  9. Location Prediction (SDM)
  10. Region Prediction using Region Proposal Network (RPN)
  11. Object Identification
  12. Class-specific Region Purification
  13. Region Alignment Network (RAN)
  14. Experiments
  15. Pre-training for RAN before being Employed to FSOD-AO
  16. ...and 11 more sections

Figures (3)

  • Figure 1: The framework of the proposed method.
  • Figure 2: The comparisons of the mean values, maximal values, minimum values and variance range between $IoU_{RPN}$ and $\widetilde{IoU}_{RAN}$ every 0.2 intervals.
  • Figure 3: Some COCO images' visualization results before and after RAN process, where the red, blue and yellow boxes are the regions of $P_{r}^{RPN}$, $\widetilde{P}_{r}^{RAN}$, and $P_{\widetilde{r}}^{GT}$ respectively.