Deep active learning for object detection

Liu, Ying; Fan, Binbin; Zhang, Weiping; Ding, Weiping; Yin, Jianwei

doi:10.1016/j.ins.2021.08.019

Cited by 42 publications

(18 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Active learning is an approach that in itself does not improve the actual annotation but can aid by determining which set of images should be annotated next in order to more efficiently improve the model. This can be achieved by determining the uncertainty on a number of unlabelled image and prioritising them for annotation [27,28]. Another alternative is weak supervision, which takes lower quality labels and is able to transfer this knowledge into the training.…”

Section: Related Workmentioning

confidence: 99%

The Challenge of Data Annotation in Deep Learning—A Case Study on Whole Plant Corn Silage

Rasmussen

Kirk²,

Moeslund

2022

Sensors

View full text Add to dashboard Cite

Recent advances in computer vision are primarily driven by the usage of deep learning, which is known to require large amounts of data, and creating datasets for this purpose is not a trivial task. Larger benchmark datasets often have detailed processes with multiple stages and users with different roles during annotation. However, this can be difficult to implement in smaller projects where resources can be limited. Therefore, in this work we present our processes for creating an image dataset for kernel fragmentation and stover overlengths in Whole Plant Corn Silage. This includes the guidelines for annotating object instances in respective classes and statistics of gathered annotations. Given the challenging image conditions, where objects are present in large amounts of occlusion and clutter, the datasets appear appropriate for training models. However, we experience annotator inconsistency, which can hamper evaluation. Based on this we argue the importance of having an evaluation form independent of the manual annotation where we evaluate our models with physically based sieving metrics. Additionally, instead of the traditional time-consuming manual annotation approach, we evaluate Semi-Supervised Learning as an alternative, showing competitive results while requiring fewer annotations. Specifically, given a relatively large supervised set of around 1400 images we can improve the Average Precision by a number of percentage points. Additionally, we show a significantly large improvement when using an extremely small set of just over 100 images, with over 3× in Average Precision and up to 20 percentage points when estimating the quality.

show abstract

Section: Related Workmentioning

confidence: 99%

The Challenge of Data Annotation in Deep Learning—A Case Study on Whole Plant Corn Silage

Rasmussen

Kirk²,

Moeslund

2022

Sensors

View full text Add to dashboard Cite

show abstract

“…Consequently, energy assessors can augment the accuracy and efficiency of energy assessments, thereby ultimately advancing the energy performance of buildings and mitigating carbon emissions. Active learning 18 addresses the challenge of obtaining labelled data by allowing machine learning algorithms to query users interactively for desired outputs, thereby minimising user intervention while achieving more robust annotation tasks and improving model accuracy. In this paper, we propose a progressive active learning workflow that has been trained on a limited number of meticulously annotated images (500 floorplan images) to iteratively rectify the annotation of 4,500 intricate images.…”

Section: Introductionmentioning

confidence: 99%

The Power of Progressive Active Learning in Floorplan Images for Energy Assessment

Al-Turki,

Kyriakou,

Basurra

et al. 2023

Preprint

View full text Add to dashboard Cite

Floorplan energy assessments present a highly efficient method for evaluating the energy efficiency of residential properties without requiring physical presence. By employing computer modeling, accurate determination of the building’s heat loss or gain can be achieved, enabling planners and homeowners to devise energy-efficient renovation or redevelopment plans. However, the creation of an AI model for floorplan element detection necessitates the manual annotation of a substantial collection of floorplans, which poses a daunting task. This paper introduces a novel active learning model designed to detect and annotate the primary elements within floorplan images, aiming to assist energy assessors in automating the analysis of such images—an inherently challenging problem due to the time-intensive nature of the annotation process. Our active learning approach initially trained on a set of 500 annotated images and progressively learned from a larger dataset comprising 4,500 unlabeled images. This iterative process resulted in an mean average precision score of 0.833, precision score of 0.972, and recall score of 0.950. The annotated dataset will be made publicly available under a Creative Commons license.

show abstract

“…[4] How to effectively exploit these unlabelled samples to improve the diagnostic model has become a meaningful topic. [5] In the face of the insufficient labelled data problem, active lLearning (AL) [6][7][8] provides a new solution to this challenge, and has been widely studied and applied to industrial fault diagnosis. AL can query information sources interactively: during each iteration, more valuable unlabelled learning samples are selected and submitted to experts for annotation.…”

Section: Introductionmentioning

confidence: 99%

A novel density ratio‐based batch active learning fault diagnosis method integrated with adaptive Laplacian graph trimming

Jiang

Zhu

et al. 2023

Can J Chem Eng

View full text Add to dashboard Cite

In actual industrial processes, although a large number of original data are easy to obtain, only a few samples are effectively labelled, which is insufficient to construct a supervised fault diagnostic model. Facing the industrial demand of fault diagnosis, in this paper, a novel density ratio (DR)‐based batch active learning (BAL) fault diagnosis method integrated with adaptive Laplacian graph trimming (ALGT) method is proposed. First, under the active learning framework, a new index DR‐based on local reachability density (LRD) is proposed to search the low density and high uncertainty samples, in which the local outliers factor (LOF) is used to search the samples in low density region and the ratio of LRD and intra‐class LRD is calculated to search the samples with high uncertainty. Second, the samples are selected and manually labelled in batches according to the proposed index DR, and the labelled data set and the unlabelled data set are updated and reconstructed. Third, based on the reconstructed labelled dataset and remaining unlabelled dataset, a semi‐supervised classifier ALGT is constructed for fault diagnosis. In ALGT, the Laplacian weighted graph is initialized and iteratively optimized by ALGT. Finally, the proposed DR‐based BAL‐ALGT (DRBAL‐ALGT) fault diagnosis method is verified by the Tennessee Eastman process (TEP) and applied to grid‐connected photovoltaic systems (GPVS). The experimental results show that the proposed DRBAL‐ALGT method can achieve higher accuracy for fault diagnosis.

show abstract

Deep active learning for object detection

Cited by 42 publications

References 12 publications

The Challenge of Data Annotation in Deep Learning—A Case Study on Whole Plant Corn Silage

The Challenge of Data Annotation in Deep Learning—A Case Study on Whole Plant Corn Silage

The Power of Progressive Active Learning in Floorplan Images for Energy Assessment

A novel density ratio‐based batch active learning fault diagnosis method integrated with adaptive Laplacian graph trimming

Contact Info

Product

Resources

About