Gap-KD: Bridging the Significant Capacity Gap Between Teacher and Student Model

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 435 - 453

Published: Jan. 1, 2025

Language: Английский

---

Multilevel parallel attention knowledge distillation for multimodal biometric recognition

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 154, P. 110865 - 110865

Published: April 29, 2025

---

SVD-KD: SVD-based hidden layer feature extraction for Knowledge distillation

Pattern Recognition, Journal Year: 2025, Volume and Issue: unknown, P. 111721 - 111721

Published: May 1, 2025

Language: Английский

---

基于DETR和改进去噪训练的光学遥感图像多尺度旋转目标检测（特邀）

Laser & Optoelectronics Progress, Journal Year: 2024, Volume and Issue: 61(2), P. 0211023 - 0211023

Published: Jan. 1, 2024

旋转目标检测是遥感图像解译的重要任务之一,存在目标方向任意、小目标密集排列、目标表示引起的角度周期性等典型问题。针对上述问题,提出一种基于DEtection Transformer(DETR)目标检测器和改进去噪训练的旋转目标检测方法,即arbitrary-oriented object detection Transformer with improved deNoising anchor boxes(AO2DINO)。首先,该方法引入一种多尺度旋转可变形注意力模块,将角度信息以旋转矩阵的形式引入注意力权重的计算,提高了模型对旋转目标的适应能力。其次,针对小目标密集排列问题,提出一种自适应的样本分配器,引入旋转交并比和自适应阈值,实现对密集目标更加精确的采样,提升了模型对小目标的检测能力。最后,在模型中引入基于卡尔曼滤波的交并比(KFIoU)作为回归损失,以解决旋转目标表示引起的角度周期性问题。AO2DINO在DOTAv1.0和DIOR-R两个公开数据集上与典型的旋转目标检测方法进行了比较,在DETR系列旋转目标检测方法中检测精度最高,且训练时收敛速度更快,在训练12个epochs时就几乎达到了其他旋转目标检测方法训练36个epochs时的检测效果。

---

Cross-domain visual prompting with spatial proximity knowledge distillation for histological image classification

Journal of Biomedical Informatics, Journal Year: 2024, Volume and Issue: 158, P. 104728 - 104728

Published: Sept. 21, 2024

Language: Английский

---