TS-BiT: Two-Stage Binary Transformer for ORSI Salient Object Detection
- 18 February 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 22 (1545598X) , 1-5
Abstract
Vision Transformers (ViTs) have demonstrated superior performance in various remote sensing tasks, such as Optical Remote Sensing Images Salient Object Detection (ORSI-SOD). However, the high resolution of remote sensing images and the substantial computational costs pose significant challenges for deploying existing methods on resource-constrained devices. Model binarization significantly reduces computational costs and storage requirements by constraining weights and activations to 1-bit representations, which has been widely explored in Convolutional Neural Networks (CNNs). However, directly applying binary methods to ViTs poses challenges since quantization errors hinder the ability to capture the similarity between tokens, resulting in significant performance degradation in detecting salient objects in complex ORSI scenarios. To address this issue, we propose TS-BiT for the ORSI-SOD task to preserve information of salient objects under 1-bit representation. Specifically, we design a two-stage central-aware softmax binarization strategy (TCSB) to reduce quantization errors arising from substantial discrepancies in the long-tail distribution of multi-head attention. Furthermore, we develop a scalable hyperbolic tangent function to approximate the gradients of the Sign function within each binarization group, substantially mitigating quantization errors during the binarization of softmax attention. Extensive experiments demonstrate that our method outperforms existing binary ViT approaches on ORSSD, EORSSD, and ORSI-4199 datasets.Keywords
Funding Information
- National Natural Science Foundation of China (61921001)
- National Natural Science Foundation of China (62022091)
This publication has 10 references indexed in Scilit:
- A Causal Adjustment Module for Debiasing Scene Graph GenerationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2025
- SARATR-X: Toward Building a Foundation Model for SAR Target RecognitionIEEE Transactions on Image Processing, 2025
- Multistage Enhancement Network for Tiny Object Detection in Remote Sensing ImagesIEEE Transactions on Geoscience and Remote Sensing, 2024
- Feature-Based Knowledge Distillation for Infrared Small Target DetectionIEEE Geoscience and Remote Sensing Letters, 2024
- BiViT: Extremely Compressed Binary Vision Transformers2023
- High-Quality Angle Prediction for Oriented Object Detection in Remote Sensing ImagesIEEE Transactions on Geoscience and Remote Sensing, 2023
- A Low Coupling and Lightweight Algorithm for Ship Detection in Optical Remote Sensing ImagesIEEE Geoscience and Remote Sensing Letters, 2022
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation2022
- Binary Neural Network for Multispectral Image ClassificationIEEE Geoscience and Remote Sensing Letters, 2022
- XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks2016