lenscloth/RKD
PyTorch implementation of CVPR 2019 Relational Knowledge Distillation for compressing deep learning models via teacher-student training.
Not currently ranked — collecting fresh signals.
star history
This repository implements Relational Knowledge Distillation, a technique for transferring structured knowledge from a larger teacher neural network to a smaller student network. The code supports metric learning experiments using architectures like ResNet, with distance-weighted sampling and triplet loss for training. It includes scripts for training teacher embeddings (e.g., ResNet50), evaluating performance, and distilling knowledge to student networks (e.g., ResNet18) for computer vision tasks like CUB-200.
Frequently asked
- What is lenscloth/RKD?
- PyTorch implementation of CVPR 2019 Relational Knowledge Distillation for compressing deep learning models via teacher-student training.
- Is RKD open source?
- Yes — lenscloth/RKD is an open-source project tracked on heatdrop.
- What language is RKD written in?
- lenscloth/RKD is primarily written in Python.
- How popular is RKD?
- lenscloth/RKD has 419 stars on GitHub.
- Where can I find RKD?
- lenscloth/RKD is on GitHub at https://github.com/lenscloth/RKD.