规模混合集团蒸与知识解用于持续的语义细分
Zichen Song1, Xiaoliang Zhang1, Zhaofeng Shi1
1School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China.
Sensors (Basel, Switzerland)
|September 28, 2023
概括
本研究引入了一种新的连续语义细分 (CSS) 方法,使用尺度混合蒸和知识解. 它可以增强学习新的视觉类别,同时保留旧的视觉类别,提高模型的稳定性和可塑性.
科学领域:
- 计算机视觉 计算机视觉
- 机器学习 机器学习
- 人工智能的人工智能
背景情况:
- 持续语义细分 (CSS) 使模型能够顺序地学习新的视觉类别,而不会忘记以前的视觉类别.
- 由于知识蒸的局限性,现有的CSS方法经常因知识利用不足和背景语义转移而扎.
研究的目的:
- 提出一种新的CSS方法,解决当前知识蒸技术的局限性.
- 提高模型的稳定性和可塑性,用于语义细分的持续学习场景.
主要方法:
- 编码器的一种规模混合组语义蒸 (SGD) 方法,通过组聚精细化转移多个规模的知识.
- 解码器的知识解蒸 (KDD) 方法,使用旧类区域指导特征地图蒸,以减少语义转移.
主要成果:
- 提出的方法证明了与Pascal VOC和ADE20K数据集的最先进方法相比具有竞争力的性能.
- 实验结果验证了规模混合蒸和知识解在提高CSS性能方面的有效性.
结论:
- 新的CCS方法有效地平衡了学习新类别和保存旧类别.
- 拟议的SGD和KDD技术为持续的语义细分提供了更强大,更有效的方法.
相关概念视频
01:08Size-Exclusion Chromatography
635
In size-exclusion chromatography (SEC), also known as molecular-exclusion or gel-permeation chromatography, molecules are separated based on their sizes. This technique is important for separating large molecules such as polymers and biomolecules. The two classes of micron-sized stationary phases encountered in SEC are silica particles and cross-linked polymer resin beads. Both materials are porous, but their pore sizes vary significantly.
Silica particles offer advantages such as rigidity,...
Silica particles offer advantages such as rigidity,...
635
01:01Distillation: Vapor–Liquid Equilibria
2.8K
Distillation is a separation technique that takes advantage of the boiling point properties of disparate elements in a mixture. To perform distillation, we begin by heating a miscible mixture of two liquids with a significant difference in boiling points (at least 20°C). As the solution heats up and reaches the bubble point of the more volatile component, some molecules of the more volatile component transition into the gas phase and travel upward into the condenser, which is a glass tube...
2.8K