Region-adaptive and context-complementary cross modulation for RGB-T semantic segmentation

Article Computer Science, Artificial Intelligence

SATS: Self-attention transfer for continual semantic segmentation

Yiqiao Qiu et al.

Summary: In this study, a new knowledge transfer method was proposed to alleviate the catastrophic forgetting issue in continuous semantic segmentation. This method captures the relationships between elements within each image using self-attention maps in a Transformer-style segmentation model. Extensive evaluations show that the proposed method outperforms state-of-the-art solutions when combined with widely adopted strategies.

PATTERN RECOGNITION (2023)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Slimmable Dataset Condensation

Songhua Liu et al.

Summary: In this paper, the authors propose a slimmable dataset condensation method that extracts a smaller synthetic dataset based on previous condensation results. They study the limitations of existing methods in a successive compression setting and identify two key factors. To address these issues, they introduce a novel training objective and a significance-aware parameterization method. Extensive comparisons and experiments demonstrate the superiority of their method over existing methods.

2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR (2023)

Add to Collection

Article Computer Science, Artificial Intelligence

Contextual ensemble network for semantic segmentation

Quan Zhou et al.

Summary: This paper introduces a novel encoder-decoder architecture called CENet for semantic segmentation, which achieves superior performance on two widely-used semantic segmentation datasets and obtains promising results on instance segmentation and biological segmentation tasks.

PATTERN RECOGNITION (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

A cascaded nested network for 3T brain MR image segmentation guided by 7T labeling *

Jie Wei et al.

Summary: A cascaded nested network (CaNes-Net) is proposed for brain MR image segmentation at 3T, trained by tissue labels delineated from 7T images, reducing segmentation errors caused by misalignment and improving accuracy substantially.

PATTERN RECOGNITION (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

Complementarity-aware cross-modal feature fusion network for RGB-T semantic segmentation

Wei Wu et al.

Summary: RGB-T semantic segmentation has gained attention for its robustness under challenging illumination. This paper proposes a Complementarity-aware Cross-modal Feature Fusion Network (CCFFNet) that selects and fuses complementary information from RGB and thermal features. Experimental results show that the proposed model outperforms state-of-the-art models and can be easily applied to multi-modal semantic segmentation.

PATTERN RECOGNITION (2022)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

CMT-DeepLab: Clustering Mask Transformers for Panoptic Segmentation

Qihang Yu et al.

Summary: This paper presents a transformer-based framework for panoptic segmentation, called Clustering Mask Transformer (CMT-DeepLab), which utilizes a CMT layer to compute pixel clustering based on feature affinity and generate denser and more consistent cross-attention for the final segmentation task. Experimental results show that this method significantly improves performance compared to prior art.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022) (2022)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Masked-attention Mask Transformer for Universal Image Segmentation

Bowen Cheng et al.

Summary: Masked-attention Mask Transformer (Mask2Former) is a new architecture capable of addressing any image segmentation task and outperforms specialized architectures on multiple datasets.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022) (2022)

Add to Collection

Article Computer Science, Information Systems

MFFENet: Multiscale Feature Fusion and Enhancement Network For RGB-Thermal Urban Road Scene Parsing

Wujie Zhou et al.

Summary: In this paper, a novel multiscale feature fusion and enhancement network (MFFENet) is proposed for accurate parsing of RGB-thermal urban road scenes. By incorporating multi-label supervision and a spatial attention mechanism module, the MFFENet outperforms similar high-performing methods and emphasizes foreground objects in the scene.

IEEE TRANSACTIONS ON MULTIMEDIA (2022)

Add to Collection

Article Automation & Control Systems

FuseSeg: Semantic Segmentation of Urban Scenes Based on RGB and Thermal Data Fusion

Yuxiang Sun et al.

Summary: Semantic segmentation of urban scenes is crucial for autonomous driving applications, and recent advancements in deep learning have led to improved performance. However, traditional networks using single-modal sensory data may struggle in challenging lighting conditions. This article introduces the FuseSeg network, which fuses RGB and thermal data to achieve superior segmentation performance in urban scenes. The network outperforms state-of-the-art networks and can be easily implemented using various deep learning frameworks.

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Attention fusion network for multi-spectral semantic segmentation

Jiangtao Xu et al.

Summary: The proposed deep learning model, AFNet, utilizes attention mechanism to calculate spatial correlation between features from different spectra, improving the accuracy and visual definition of multi-spectral semantic segmentation.

PATTERN RECOGNITION LETTERS (2021)

Add to Collection

Article Computer Science, Artificial Intelligence