Abstract:Current methods for low- and few-shot object detection have primarily focused on enhancing model performance for detecting objects. One common approach to achieve this is by combining model finetuning with data augmentation strategies. However, little attention has been given to the energy efficiency of these approaches in data-scarce regimes. This paper seeks to conduct a comprehensive empirical study that examines both model performance and energy efficiency of custom data augmentations and automated data augmentation selection strategies when combined with a lightweight object detector. The methods are evaluated in three different benchmark datasets in terms of their performance and energy consumption, and the Efficiency Factor is employed to gain insights into their effectiveness considering both performance and efficiency. Consequently, it is shown that in many cases, the performance gains of data augmentation strategies are overshadowed by their increased energy usage, necessitating the development of more energy efficient data augmentation strategies to address data scarcity.

What problem does this paper attempt to address?

The paper primarily aims to address the impact of data augmentation strategies on model performance and energy consumption in Low-Shot Learning (LSL) and Few-Shot Learning (FSL) object detection under data scarcity conditions. Specifically: 1. **Research Background**: In industrial applications, training machine learning and deep learning models typically requires a large amount of data to improve model performance. However, in certain fields (such as industry or healthcare), collecting large amounts of data can be costly or pose privacy protection issues. Therefore, Data Augmentation (DA) techniques are widely used to address data scarcity problems. 2. **Main Issue**: Although existing research mainly focuses on how to improve model performance through data augmentation, little attention has been paid to the energy consumption of these methods in data-scarce scenarios. The paper points out that in practical applications, especially when running models on resource-constrained edge devices, energy consumption is a critical factor. 3. **Research Objective**: This paper aims to experimentally evaluate the performance and energy consumption of different data augmentation strategies in low-shot and few-shot object detection tasks and proposes an Efficiency Factor indicator that comprehensively considers model performance and energy consumption. Specifically, the study uses three benchmark datasets for evaluation, including datasets for Personal Protective Equipment (PPE), Fire scenes, and Construction Sites. 4. **Summary of Contributions**: - Analyzed the performance and energy consumption of various data augmentation strategies in low-shot and few-shot object detection tasks. - Conducted ablation experiments to compare the differences between custom data augmentation methods and automated data augmentation selection methods. - For the first time, proposed a customized Efficiency Factor indicator to evaluate the overall efficiency of data augmentation methods in low-shot and few-shot learning. In summary, this paper aims to fill the gap in existing research regarding the energy consumption evaluation of data augmentation strategies in low-shot and few-shot object detection tasks, with the goal of developing more energy-efficient data augmentation methods.

A Closer Look at Data Augmentation Strategies for Finetuning-Based Low/Few-Shot Object Detection

Evaluating the Energy Efficiency of Few-Shot Learning for Object Detection in Industrial Settings

Meta-tuning Loss Functions and Data Augmentation for Few-shot Object Detection

Using Data Augmentation to Improve the Generalization Capability of an Object Detector on Remote-Sensed Insect Trap Images

Enhancing Monocular 3-D Object Detection Through Data Augmentation Strategies

Optimizing Few-Shot Remote Sensing Scene Classification Based on an Improved Data Augmentation Approach

How Important are Data Augmentations to Close the Domain Gap for Object Detection in Orbit?

Data-Efficient Augmentation for Training Neural Networks

Scale-aware Automatic Augmentations for Object Detection with Dynamic Training

Data Augmentation Based on Cannikin Law for Object Detection

Data augmentation and few-shot change detection in forest remote sensing

Leveraging the Power of Data Augmentation for Transformer-based Tracking

Optimizing Automated Optical Inspection: An Adaptive Fusion and Semi-Supervised Self-Learning Approach for Elevated Accuracy and Efficiency in Scenarios with Scarce Labeled Data

Optimizing distortion magnitude for data augmentation in few-shot remote sensing scene classification

Randaugment: Practical automated data augmentation with a reduced search space

What makes a good data augmentation for few-shot unsupervised image anomaly detection?

Simple In-place Data Augmentation for Surveillance Object Detection

Improving 3D Object Detection through Progressive Population Based Augmentation

Comparative Analysis of Data Augmentation for Retinal OCT Biomarker Segmentation

Quantifying Data Augmentation for LiDAR based 3D Object Detection

Augmentation for small object detection