UCP-Net: Unstructured Contour Points for Instance Segmentation

Dupont, Camille; Ouakrim, Yanis; Pham, Quoc Cuong

doi:10.1109/smc52423.2021.9658754

Cited by 11 publications

(2 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For deep segmentation techniques, exploitation of the boundary has been realized by introducing bounding boxes [13–15]. For example, Deep Extreme Cut (DEXTR) [13] requires users to provide four extreme points on the target boundary to facilitate segmentation using Convolutional Neural Networks (CNN).…”

Section: Related Workmentioning

confidence: 99%

TriClick: Interactive Dermoscopic Image Segmentation with Triangle Map

Zhao,

Gu,

et al. 2024

IEEJ Transactions Elec Engng

View full text Add to dashboard Cite

Lesion segmentation is a fundamental task that has been widely studied in biomedicine. State‐of‐the‐art methods still struggle with the inherent challenges of dermoscopic images such as uncertain boundaries or severe occlusion. In this study, we argue that this task can be better resolved by introducing user interaction and proposing a deep interactive framework for lesion segmentation. Our method is designed to best reflect user expertise to iteratively refine the segmentation results. In each iteration, professional users are expected to click three points on the boundary of the lesion to encourage the network to update the previous round of the segmentation result. To best exploit the clues from the interaction, we explore a novel encoding strategy by producing a triangle map whose vertices are the three clicked points. Owing to the strong guidance within such triangle maps, our method can outperform state‐of‐the‐art methods with baseline segmentation backbones. Extensive experimental results and ablation studies on three commonly used lesion segmentation data sets demonstrated the effectiveness of our method in terms of segmentation accuracy and encoding effectiveness. © 2024 The Authors. IEEJ Transactions on Electrical and Electronic Engineering published by Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

show abstract

Section: Related Workmentioning

confidence: 99%

TriClick: Interactive Dermoscopic Image Segmentation with Triangle Map

Zhao,

Gu,

et al. 2024

IEEJ Transactions Elec Engng

View full text Add to dashboard Cite

show abstract

“…On the other hand, traditional interactive building extraction methods for sample-free remote sensing images only rely on manual fine tracing or clicking to complete fine building extraction, which has greater adaptability compared to fully automated methods. However, traditional interactive segmentation methods are based on hand-crafted features, meaning that the pixel-level labeling of natural images takes an average of 10.1 min per image [16][17][18][19][20][21][22], while remote sensing image annotation is even more time-consuming and challenging due to various factors, such as water vapor and lighting conditions [23,24]. Hence, it is vital to utilize deep learning techniques to understand objects and semantics and achieve the high-precision extraction of structures at a lower cost.…”

Section: Introductionmentioning

confidence: 99%

ARE-Net: An Improved Interactive Model for Accurate Building Extraction in High-Resolution Remote Sensing Imagery

Weng,

Wang,

Lin

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Accurate building extraction for high-resolution remote sensing images is critical for topographic mapping, urban planning, and many other applications. Its main task is to label each pixel point as a building or non-building. Although deep-learning-based algorithms have significantly enhanced the accuracy of building extraction, fully automated methods for building extraction are limited by the requirement for a large number of annotated samples, resulting in a limited generalization ability, easy misclassification in complex remote sensing images, and higher costs due to the need for a large number of annotated samples. To address these challenges, this paper proposes an improved interactive building extraction model, ARE-Net, which adopts a deep interactive segmentation approach. In this paper, we present several key contributions. Firstly, an adaptive-radius encoding (ARE) module was designed to optimize the interaction features of clicks based on the varying shapes and distributions of buildings to provide maximum a priori information for building extraction. Secondly, a two-stage training strategy was proposed to enhance the convergence speed and efficiency of the segmentation process. Finally, some comprehensive experiments using two models of different sizes (HRNet18s+OCR and HRNet32+OCR) were conducted on the Inria and WHU building datasets. The results showed significant improvements over the current state-of-the-art method in terms of NoC90. The proposed method achieved performance enhancements of 7.98% and 13.03% with HRNet18s+OCR and 7.34% and 15.49% with HRNet32+OCR on the WHU and Inria datasets, respectively. Furthermore, the experiments demonstrated that the proposed ARE-Net method significantly reduced the annotation costs while improving the convergence speed and generalization performance.

show abstract