A full data augmentation pipeline for small object detection based on generative adversarial networks

Bosquet, Brais; Cores, Daniel; Seidenari, Lorenzo; Brea, V.M.; Mucientes, Manuel; Bimbo, Alberto Del

doi:10.1016/j.patcog.2022.108998

Cited by 44 publications

(14 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Various methods have been developed and tested to balance dataset upsampling for object detection tasks [ 9 , 10 , 11 , 12 , 13 ]. However, additional difficulties arise due to the need to properly modify outputs, such as bounding boxes, when applying transformations to input images.…”

Section: Proposalmentioning

confidence: 99%

POSEIDON: A Data Augmentation Tool for Small Object Detection Datasets in Maritime Environments

Ruiz-Ponce

Ortiz-Perez

García-Rodríguez

et al. 2023

Sensors

View full text Add to dashboard Cite

Certain fields present significant challenges when attempting to train complex Deep Learning architectures, particularly when the available datasets are limited and imbalanced. Real-time object detection in maritime environments using aerial images is a notable example. Although SeaDronesSee is the most extensive and complete dataset for this task, it suffers from significant class imbalance. To address this issue, we present POSEIDON, a data augmentation tool specifically designed for object detection datasets. Our approach generates new training samples by combining objects and samples from the original training set while utilizing the image metadata to make informed decisions. We evaluate our method using YOLOv5 and YOLOv8 and demonstrate its superiority over other balancing techniques, such as error weighting, by an overall improvement of 2.33% and 4.6%, respectively.

show abstract

Section: Proposalmentioning

confidence: 99%

POSEIDON: A Data Augmentation Tool for Small Object Detection Datasets in Maritime Environments

Ruiz-Ponce

Ortiz-Perez

García-Rodríguez

et al. 2023

Sensors

View full text Add to dashboard Cite

show abstract

“…To address this issue, researchers have proposed several methods that fall into four main categories: multiscale feature learning, 39,40 improving input feature resolution, 9,41 integration context knowledge, 42,43 and data augmentation. 44,45 Multiscale feature learning involves combining low-level spatial information with high-level semantic information to improve object features. For example, Gong et al 10 added fusion factors between adjacent layers in FPN to help the shallow layers focus on learning small object features, which leads to improvement for small object detection.…”

Section: Small Object Detection On the Highwaymentioning

confidence: 99%

“…Autonomous vehicles often struggle to detect small object features on mountainous highways. To address this issue, researchers have proposed several methods that fall into four main categories: multiscale feature learning, 39 , 40 improving input feature resolution, 9 , 41 integration context knowledge, 42 , 43 and data augmentation 44 , 45 . Multiscale feature learning involves combining low-level spatial information with high-level semantic information to improve object features.…”

Section: Related Workmentioning

confidence: 99%

Small object detection for autonomous driving under hazy conditions on mountain motorways

Wang,

Qiu,

Jiang

et al. 2023

Opt. Eng.

View full text Add to dashboard Cite

To address the issue of high miss rates for distant small objects and the diminished system detection performance due to the influence of hazy when autonomous vehicles operate on mountain highways. We propose a framework for small object vehicle detection in hazy traffic environments (SHTDet). This framework aims to enhance small object detection for autonomous driving under hazy conditions on mountainous motorways. Specifically, to restore the clarity of hazy images, we designed an image enhancement (IE), and its parameters are predicted by a convolutional neural network [filter parameter estimation (FPE)]. In addition, to enhance the detection accuracy of small objects, we introduce a cascaded sparse query (CSQ) mechanism, which effectively utilizes high-resolution features while maintaining fast detection speed. We jointly optimize the IE and the detection network (CSQ-FCOS) in an end-to-end manner, ensuring that FPE module can learn a suitable IE. Our proposed SHTDet method is adept at adaptively handling sunny and hazy conditions. Extensive experiments demonstrate the efficacy of the SHTDet method in detecting small objects on hazy sections of mountain highways.

show abstract

“…(Yan et al, 2019) The dataset trained a faster region-based convolutional neural network (Faster R-CNN) built on Res101netwok, which was then used to classify both synthetic and real images. (Bosquet et al, 2022) Synthetic data of superior quality achieved by combining a GAN with image inpainting and mixing. DS-GAN can create believable miniature things.…”

Section: Deep Learning-based Object Detection Algorithm Improvementmentioning

confidence: 99%

AI-based object detection latest trends in remote sensing, multimedia and agriculture applications

Nawaz

Bhatti

et al. 2022

Front. Plant Sci.

View full text Add to dashboard Cite

Object detection is a vital research direction in machine vision and deep learning. The object detection technique based on deep understanding has achieved tremendous progress in feature extraction, image representation, classification, and recognition in recent years, due to this rapid growth of deep learning theory and technology. Scholars have proposed a series of methods for the object detection algorithm as well as improvements in data processing, network structure, loss function, and so on. In this paper, we introduce the characteristics of standard datasets and critical parameters of performance index evaluation, as well as the network structure and implementation methods of two-stage, single-stage, and other improved algorithms that are compared and analyzed. The latest improvement ideas of typical object detection algorithms based on deep learning are discussed and reached, from data enhancement, a priori box selection, network model construction, prediction box selection, and loss calculation. Finally, combined with the existing challenges, the future research direction of typical object detection algorithms is surveyed.

show abstract

A full data augmentation pipeline for small object detection based on generative adversarial networks

Cited by 44 publications

References 13 publications

POSEIDON: A Data Augmentation Tool for Small Object Detection Datasets in Maritime Environments

POSEIDON: A Data Augmentation Tool for Small Object Detection Datasets in Maritime Environments

Small object detection for autonomous driving under hazy conditions on mountain motorways

AI-based object detection latest trends in remote sensing, multimedia and agriculture applications

Contact Info

Product

Resources

About