GeorgeCazenavette/mtt-distillation
Official code for our CVPR '22 paper "Dataset Distillation by Matching Training Trajectories"
This project helps machine learning researchers and practitioners create extremely small, synthetic image datasets that can train models to perform as well as models trained on much larger, real datasets. You input a large, real image dataset, and it outputs a tiny collection of synthetic images. This is useful for anyone working with image classification who needs to reduce training time or storage requirements for their models.
438 stars. No commits in the last 6 months.
Use this if you need to drastically shrink the size of an image dataset while preserving its ability to train high-performing machine learning models.
Not ideal if your primary goal is to improve model accuracy beyond what the original full dataset allows, or if you are working with non-image data.
Stars
438
Forks
63
Language
Python
License
—
Category
Last pushed
Jul 16, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/GeorgeCazenavette/mtt-distillation"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.