GeorgeCazenavette/mtt-distillation
Research code for Dataset Distillation by Matching Training Trajectories, which learns synthetic images to replace real datasets for training computer vision models.
Not currently ranked — collecting fresh signals.
star history
This repository implements a dataset distillation method published at CVPR 2022. The approach learns a small set of synthetic images such that models trained exclusively on them achieve similar test performance to models trained on the full real dataset. It works by optimizing synthetic data to induce similar training dynamics in student networks as expert networks trained on real data, measuring error in parameter space across training iterations.
Frequently asked
- What is GeorgeCazenavette/mtt-distillation?
- Research code for Dataset Distillation by Matching Training Trajectories, which learns synthetic images to replace real datasets for training computer vision models.
- Is mtt-distillation open source?
- Yes — GeorgeCazenavette/mtt-distillation is an open-source project tracked on heatdrop.
- What language is mtt-distillation written in?
- GeorgeCazenavette/mtt-distillation is primarily written in Python.
- How popular is mtt-distillation?
- GeorgeCazenavette/mtt-distillation has 440 stars on GitHub.
- Where can I find mtt-distillation?
- GeorgeCazenavette/mtt-distillation is on GitHub at https://github.com/GeorgeCazenavette/mtt-distillation.