Projects sharing features with Mtt Distillation

This is the official repository for paper Discovering Galaxy Features via Dataset Distillation. Our work contains the experiment code for galaxy dataset distillation and Self-Adaptive Trajectory Matching (STM) algorithm, an improved version of Matching Training Trajectory (MTT).

Existing DD methods exhibit degraded performance when applied to imbalanced datasets, especially when the imbalance factor increases, whereas our method provides significantly better performance under different imbalanced scenarios.

Hello!!! Thanks for checking out our repo and paper! 🍻

The official implementation of A Large-Scale Study on Video Action Dataset Condensation

This repository contains code for training expert trajectories and distilling synthetic data for the paper: Dataset Distillation by Automatic Training Trajectories. The listed is the steps to run the code. 1. Set up enveriments. 2. Create an wandb account for monitoring distillation process…

In this work, we propose to emphasize discriminative features for dataset distillation in the complex scenario, i.e. images in complex scenarios are characterized by significant variations in object sizes and the presence of a large amount of class-irrelevant information.

Matching-based Dataset Distillation methods can be summarized into two steps:

This repository is the official implementation of Dataset Condensation with Contrastive Signals (DCC), published as a conference paper at ICML 2022. The implementation is based on (https://github.com/VICO-UoE/DatasetCondensation).

Paper

Official PyTorch implementation of "Loss-Curvature Matching for Dataset Selection and Condensation" (AISTATS 2023) by Seungjae Shin, HeeSun Bae, Donghyeok Shin, Weonyoung Joo, and Il-Chul Moon.

PyTorch implementation of paper "Neural Spectral Decomposition for Dataset Distillation" in ECCV 2024.

This repository utilizes PyTorch and modern experiment manager tools, Hydra and Wandb.

Dataset condensation aims to condense a large training set T into a small synthetic set S such that the model trained on the small synthetic set can obtain comparable testing performance to that trained on the large training set.

Wenliang Zhong 1 , Haoyu Tang 1 , Qinghai Zheng 2 , Mingzhu Xu 1 , Yupeng Hu 1 , Weili Guan 3