DeepKD is a knowledge distillation trainer that decouples task, target-class, and non-target-class gradients with GSNR-based momentum and a dynamic top-k mask, yielding consistent accuracy gains on CIFAR-100, ImageNet, and MS-COCO.
Cross-image relational knowledge distillation for semantic segmentation,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
DeepKD: A Deeply Decoupled and Denoised Knowledge Distillation Trainer
DeepKD is a knowledge distillation trainer that decouples task, target-class, and non-target-class gradients with GSNR-based momentum and a dynamic top-k mask, yielding consistent accuracy gains on CIFAR-100, ImageNet, and MS-COCO.