Color-Oriented Redundancy Reduction in Dataset Distillation

Authors: Bowen Yuan, Zijian Wang, Mahsa Baktashmotlagh, Yadan Luo, Zi Huang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental A comprehensive performance study involving various datasets and evaluation scenarios is conducted, demonstrating the superior performance of our proposed color-aware DD compared to existing DD methods.
Researcher Affiliation Academia Bowen Yuan Zijian Wang Mahsa Baktashmotlagh Yadan Luo Zi Huang {bowen.yuan, zijian.wang, m.baktashmotlagh, y.luo, helen.huang}@uq.edu.au The University of Queensland
Pseudocode Yes Algorithm 1: Algorithm for guided image selection with maximum information gain.
Open Source Code Yes The code is available at https://github.com/Ke Vi NYu An0314/Auto Palette.
Open Datasets Yes We conduct experiments of our model on various benchmark datasets, including CIFAR-10 [21], CIFAR-100 [21] and Image Net [7]. CIFAR10: an image dataset consists of 50,000 32 32 RGB images for training, and 10,000 images for testing.
Dataset Splits No The paper does not explicitly provide training/validation/test dataset splits, only mentioning training and testing sets for CIFAR10 and total image counts for CIFAR100.
Hardware Specification Yes All experiments can be conducted on 2 Nvidia H100 GPUs that have 80GB RAM for each or 4 Nvidia V100 GPUs that have 32GB RAM for each.
Software Dependencies No The paper does not provide specific version numbers for software dependencies such as libraries, frameworks, or compilers used in the experiments.
Experiment Setup Yes We set loss coefficients α=1, β=1, γ=3 for all experiments if not specified. Table 9: Hyperparameters for our method based on Distribution matching (DM) framework. Table 10: Hyperparameters for our method based on Trajectory matching (TM) framework. These tables provide specific hyperparameters.