IDA: Improved Data Augmentation Applied to Salient Object Detection

Ruíz, Daniel; Krinski, Bruno A.; Todt, Eduardo

doi:10.1109/sibgrapi51738.2020.00036

Cited by 9 publications

(5 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the parameters of the AR images in fog condition, the transparency of the weather layer was adjusted. According to KMA's weather condition definition (22), heavy fog allows visibility less than 40 m, and it was expected to be difficult to distinguish the effect of fog on the images when the distance between the VPDS camera and vehicles is less than 5 m. There was a limitation that the difference was distinguished visually by setting the transparency lower than the expected value in consideration of the case where the distance between the VPDS and vehicles is far enough to show the fog effect.…”

Section: Discussionmentioning

confidence: 99%

“…In the augmentation process, the original label information is converted into its corresponding AR image without re-labeling. Previous studies indicate that the performance of the CNN algorithm was improved through augmentation and was helpful for generalization (20)(21)(22). To improve the performance of a large number of image classification tasks such as ImageNet ( 23), performance has been increased using simple methods such as image translations, horizontal reflections, and changing red, green, and blue (RGB) pixel values.…”

Section: Generation Of the Ar Imagesmentioning

confidence: 99%

See 1 more Smart Citation

Enhancing the Performance of Vehicle Passenger Detection under Adverse Weather Conditions Using Augmented Reality-Based Machine Learning Approach

Lee¹,

Kang²,

Lim³

et al. 2021

Transportation Research Record

View full text Add to dashboard Cite

In response to extreme traffic congestion in metropolitan areas that causes unnecessarily long travel times, high fuel consumption, and excessive greenhouse gas emissions, transportation agencies have implemented various strategies to mitigate traffic congestion. Managed lanes—one of the measures applied worldwide—provide benefits to road users and operating agencies by integrating advanced technologies such as electronic and dynamic tolling systems. However, those agencies already implementing or considering implementing the managed lane strategy are seeking a solution to effectively and properly charge toll rates based on vehicle occupancy and penalize violating vehicles. Vehicle passenger detection systems (VPDSs) have been developed and evaluated worldwide, but limitations still inhibit their full implementation. This study confirms that the performance of the deep learning algorithm, a core VPDS technology, declines under certain adverse weather conditions because of lack of training data sets. The performance of the “you only look once” (YOLOv3) model trained with a normal weather data set decreased by as much as 8.5% when it was tested for adverse weather conditions. In this study, augmented reality (AR) models are developed to enhance the accuracy of vehicle passenger detection (VPDA) by the VPDS by training the algorithm with AR images representing virtual adverse weather conditions. Models trained with AR image sets of various weather categories (fog, rain, and snow) attained VPDA enhanced by up to 7.9%. The final model significantly improves VPDA under adverse weather conditions. The proposed models could be considered for implementation with road weather information systems under adverse weather conditions.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Generation Of the Ar Imagesmentioning

confidence: 99%

Enhancing the Performance of Vehicle Passenger Detection under Adverse Weather Conditions Using Augmented Reality-Based Machine Learning Approach

Lee¹,

Kang²,

Lim³

et al. 2021

Transportation Research Record

View full text Add to dashboard Cite

show abstract

“…To verify the effectiveness of our data augmentation method, five methods are compared with ours. These methods are without data augmentation, H-Flip [14], ANDA [49], IDA [7], and GridMask [31]. In addition to the data augmentation method changes, the other parts of the network are unchanged.…”

Section: Compared With Recent Data Augmentation Methodsmentioning

confidence: 99%

“…These methods can achieve better image classification results, but cannot obtain pixel‐wise virtual knowledge. The self‐distillation method based on data augmentation [7] can increase the number and diversity of training samples, but may lose the local information between samples due to different distortion and rotation. The above problems will inevitably lead to the failure of object segmentation.…”

Section: Introductionmentioning

confidence: 99%

A self‐distillation object segmentation method via frequency domain knowledge augmentation

Chen

Cao

Zheng

et al. 2023

IET Computer Vision

View full text Add to dashboard Cite

Most self-distillation methods need complex auxiliary teacher structures and require lots of training samples in object segmentation task. To solve this challenging, a selfdistillation object segmentation method via frequency domain knowledge augmentation is proposed. Firstly, an object segmentation network which efficiently integrates multilevel features is constructed. Secondly, a pixel-wise virtual teacher generation model is proposed to drive the transferring of pixel-wise knowledge to the object segmentation network through self-distillation learning, so as to improve its generalisation ability. Finally, a frequency domain knowledge adaptive generation method is proposed to augment data, which utilise differentiable quantisation operator to adjust the learnable pixel-wise quantisation table dynamically. What's more, we reveal convolutional neural network is more inclined to learn low-frequency information during the train. Experiments on five object segmentation datasets show that the proposed method can enhance the performance of the object segmentation network effectively. The boosting performance of our method is better than recent self-distillation methods, and the average F β and mIoU are increased by about 1.5% and 3.6% compared with typical feature refinement self-distillation method.

show abstract

“…They showed that this strategy increases the number of anchor boxes generated by the Mask-RCNN [He et al 2017], which helps the network to learn and detect small objects. The ANDA [Ruiz et al 2019] and IDA [Ruiz et al 2020a] techniques follow the idea of introducing new objects, however since those are techniques focused on the generic problem of Salient Object Detection (SOD), some additional operations are necessary such as Image Inpainting to erase the original object and some additional computation to choose which combination of background and object produce a significant salience and the affine transformations to be applied to the new object that will replace the original one.…”

Section: Related Workmentioning

confidence: 99%

Light In The Black: An Evaluation of Data Augmentation Techniques for COVID-19 CT's Semantic Segmentation

Krinski¹,

Ruíz²,

Todt³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

With the COVID-19 global pandemic, computer-assisted diagnoses of medical images have gained much attention, and robust methods of Semantic Segmentation of Computed Tomography (CT) became highly desirable. Semantic Segmentation of CT is one of many research fields of automatic detection of COVID-19 and has been widely explored since the COVID-19 outbreak. In this work, we propose an extensive analysis of how different data augmentation techniques improve the training of encoder-decoder neural networks on this problem. Twenty different data augmentation techniques were evaluated on five different datasets. Each dataset was validated through a five-fold crossvalidation strategy, thus resulting in over 3,000 experiments. Our findings show that spatial level transformations are the most promising to improve the learning of neural networks on this problem.Resumo. Com a COVID-19, diagnósticos de imagens médicas assistidos por computador ganharam muita atenc ¸ão, e métodos robustos de Segmentac ¸ão Semântica de Tomografia Computadorizada (TC) tornaram-se altamente desejáveis. A Segmentac ¸ão Semântica de TC é um dos muitos campos de pesquisa de detecc ¸ão automática da COVID-19 e foi amplamente explorado desde o surto da COVID-19. Neste trabalho, propomos uma análise extensiva sobre o quanto diferentes técnicas de aumento de dados contribuem para melhorar o treinamento de redes neurais codificador-decodificador sobre este problema. Vinte técnicas diferentes de aumento de dados foram avaliadas em cinco conjuntos de dados diferentes. Cada conjunto de dados foi validado através de uma estratégia de validac ¸ão cruzada de cinco subconjuntos, resultando assim em mais de 3.000 experimentos. Nossas descobertas mostram que as transformac ¸ões de nível espacial são as mais promissoras para melhorar o aprendizado das redes neurais sobre este problema.

show abstract

IDA: Improved Data Augmentation Applied to Salient Object Detection

Cited by 9 publications

References 35 publications

Enhancing the Performance of Vehicle Passenger Detection under Adverse Weather Conditions Using Augmented Reality-Based Machine Learning Approach

Enhancing the Performance of Vehicle Passenger Detection under Adverse Weather Conditions Using Augmented Reality-Based Machine Learning Approach

A self‐distillation object segmentation method via frequency domain knowledge augmentation

Light In The Black: An Evaluation of Data Augmentation Techniques for COVID-19 CT's Semantic Segmentation

Contact Info

Product

Resources

About