Dead pixel test using effective receptive field

Kim, Bum Joon; Choi, Hyeyeon; Jang, Hyeonah; Lee, Dong Gu; Jeong, Wonseok; Kim, Sang Woo

doi:10.1016/j.patrec.2023.02.018

Cited by 12 publications

(7 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…ERF has been widely used for image classification [19,20]; however, ERF for semantic segmentation has been rarely discussed. Therefore, we first formulate an ERF for semantic segmentation.…”

Section: Erf For Semantic Segmentation Networkmentioning

confidence: 99%

See 1 more Smart Citation

Feature High-Boosting for Semantic Segmentation

Kim

Lee

et al. 2022

IEEE Access

View full text Add to dashboard Cite

Semantic segmentation has been actively studied with the great success of deep learning. Various network architectures for this task have been introduced over last few years, and the current stateof-the-art methods have shown significant advances enough to be applied to practical applications. However, previous methods often fail to segment small objects due to the limit on the spatial resolution of the feature map encoded through the deep network architecture. In this paper, we propose a simple yet effective module for semantic segmentation. Its key components include feature high-boosting, which efficiently highlights boundaries of small objects by simple operations, i.e., sum of the original feature and the corresponding residual, in the latent space. This is fairly desirable to guide network parameters to focus on features of small objects. Experimental results on various benchmarks show that the proposed method successfully improves the performance of semantic segmentation regardless of backbone networks. In particular, the proposed method with the transformer-based decoder achieves 56.4 mIoU on the ADE20K dataset.INDEX TERMS Semantic segmentation, deep learning, feature high-boosting, boundaries of small objects, latent space.

show abstract

Section: Erf For Semantic Segmentation Networkmentioning

confidence: 99%

“…The ERF of CNN is known as a symmetric 2D Gaussian [19,20]. However, understanding a scene can require information on specific regions, depending on the dataset.…”

Section: Asymmetric Pattern For Cityscapesmentioning

confidence: 99%

Feature High-Boosting for Semantic Segmentation

Kim

Lee

et al. 2022

IEEE Access

View full text Add to dashboard Cite

show abstract

“…The ERF depicts the actual usage of each pixel for determining the target feature in a neural network, representing a generic connection between them. We follow the common tricks to obtain ERFs of CNNs [12]. However, unlike CNNs, to obtain the ERF of ViT, we should focus on the patch unit.…”

Section: Introductionmentioning

confidence: 99%

“…To capture the general behavior of the ViT, G is averaged over a sufficiently large number of images. At this time, because negative values in G cancel out the positive values, we ignore the negative importance using ReLU [12][13][14]. Thus, we obtain R = 1…”

Section: Introductionmentioning

confidence: 99%

Extending class activation mapping using Gaussian receptive field

Kim

Koo

Choi

et al. 2023

Computer Vision and Image Understanding

View full text Add to dashboard Cite

“…4 and Tab. 1, we analyze the effective receptive field of LK-CSPdarknet and CSPdarknet with different input sizes[10][18][19]. We find that large kernel have larger effective receptive field in large input size.…”

mentioning

confidence: 99%

Modernized YOLOv4 with large kernels and AVOD2K dataset

Luo

Liu²,

Wang³

et al. 2023

Third International Symposium on Computer Engineering and Intelligent Communications (ISCEIC 2022)

View full text Add to dashboard Cite

We conduct research on a practical task called Armored Vehicle Object Detection (AVOD), which is designed to real-time locate and identify armored vehicles in the armored cluster. Considering the scarcity of available datasets under the background of AVOD, we carefully collect a new dataset named AVOD2K, containing 2,000 pictures which consist of nine types of armored vehicles in various environments. AVOD2K complements the missing armored vehicle dataset, and it could drive special vehicle object detection in complex scenarios. In addition, we use YOLOv4 as the baseline of the task and propose LK-CSPDarkNet to modernize YOLOv4 by combining depthwise separable convolutions and large kernels. As the parameters decreased by 7% and FLOPs a little drop, the modernized YOLOv4 outperforms the baseline in 2.4 AP.

show abstract

Dead pixel test using effective receptive field

Cited by 12 publications

References 11 publications

Feature High-Boosting for Semantic Segmentation

Feature High-Boosting for Semantic Segmentation

Extending class activation mapping using Gaussian receptive field

Modernized YOLOv4 with large kernels and AVOD2K dataset

Contact Info

Product

Resources

About