LSTD: A Low-Shot Transfer Detector for Object Detection

Abstract

Recent advances in object detection are mainly driven by deep learning with large-scale detection benchmarks. However, the fully-annotated training set is often limited for a target detection task, which may deteriorate the performance of deep detectors. To address this challenge, we propose a novel low-shot transfer detector (LSTD) in this paper, where we leverage rich source-domain knowledge to construct an effective target-domain detector with very few training examples. The main contributions are described as follows. First, we design a flexible deep architecture of LSTD to alleviate transfer difficulties in low-shot detection. This architecture can integrate the advantages of both SSD and Faster RCNN in a unified deep framework. Second, we introduce a novel regularized transfer learning framework for low-shot detection, where the transfer knowledge (TK) and background depression (BD) regularizations are proposed to leverage object knowledge respectively from source and target domains, in order to further enhance fine-tuning with a few target images. Finally, we examine our LSTD on a number of challenging low-shot detection experiments, where LSTD outperforms other state-of-the-art approaches. The results demonstrate that LSTD is a preferable deep detector for low-shot scenarios.

arXiv commentAccepted by AAAI2018

Cited by in corpus (35)

• Few-shot Object Detection on Remote Sensing Images
• Recent Advances in Object Detection in the Age of Deep Convolutional Neural Networks
• Small Sample Learning in Big Data Era
• Few-Shot Object Detection with Attention-RPN and Multi-Relation Detector
• FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding
• Restoring Negative Information in Few-Shot Object Detection
• Recent Advances in Deep Learning for Object Detection
• Meta-DETR: Image-Level Few-Shot Object Detection with Inter-Class Correlation Exploitation
• Semantic Relation Reasoning for Shot-Stable Few-Shot Object Detection
• Multi-Scale Positive Sample Refinement for Few-Shot Object Detection
• One-Shot Object Detection without Fine-Tuning
• DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection
• Cooperating RPN's Improve Few-Shot Object Detection
• Deep traffic light detection by overlaying synthetic context on arbitrary natural images
• Dynamic Relevance Learning for Few-Shot Object Detection
• A Comparative Review of Recent Few-Shot Object Detection Algorithms
• Dense Relation Distillation with Context-aware Aggregation for Few-Shot Object Detection
• Few-Shot Learning for Road Object Detection
• Hallucination Improves Few-Shot Object Detection
• FGN: Fully Guided Network for Few-Shot Instance Segmentation
• Leveraging Bottom-Up and Top-Down Attention for Few-Shot Object Detection
• Overcoming Classifier Imbalance for Long-tail Object Detection with Balanced Group Softmax
• CAT: Cross-Attention Transformer for One-Shot Object Detection
• Meta-learning algorithms for Few-Shot Computer Vision
• Boosting Weakly Supervised Object Detection with Progressive Knowledge Transfer
• Towards Generalized and Incremental Few-Shot Object Detection
• Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector
• The Devil is in Classification: A Simple Framework for Long-tail Object Detection and Instance Segmentation
• Few-shot Object Detection with Self-adaptive Attention Network for Remote Sensing Images
• Attention Guided Cosine Margin For Overcoming Class-Imbalance in Few-Shot Road Object Detection
• Meta R-CNN : Towards General Solver for Instance-level Few-shot Learning
• Meta Guided Metric Learner for Overcoming Class Confusion in Few-Shot Road Object Detection
• A Few-Shot Sequential Approach for Object Counting
• Few-shot Object Detection with Feature Attention Highlight Module in Remote Sensing Images
• Few-Shot Video Object Detection