Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

A Unified Analysis of Mixed Sample Data Augmentation: A Loss Function Perspective

Authors: Chanwoo Park, Sangdoo Yun, Sanghyuk Chun

NeurIPS 2022 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical study shows that our HMix and GMix outperform the previous state-of-the-art MSDA methods in CIFAR-100 and Image Net classification tasks.
Researcher Affiliation Collaboration Chanwoo Park , MIT EMAIL Sangdoo Yun NAVER AI Lab EMAIL Sanghyuk Chun NAVER AI Lab EMAIL
Pseudocode No The paper does not include pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Source code is available at https://github.com/naver-ai/hmix-gmix.
Open Datasets Yes Our empirical study shows that our HMix and GMix outperform the previous state-of-the-art MSDA methods in CIFAR-100 and Image Net classification tasks. ... Results on CIFAR-100 classification. ... Results on Image Net-1K classification.
Dataset Splits No The paper mentions evaluating on a "validation dataset" for Figure 5, but does not provide explicit split percentages or sizes for all datasets used in the experiments in the main text.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types) used to run its experiments in the main text.
Software Dependencies No The paper mentions software like PyTorch Image Models but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We train networks for 300 epochs using SGD optimizer with a learning rate 0.2. ... We set the hyper-parameter α for Mixup, Cut Mix, and Stochastic Mixup & Cut Mix to 1. α for HMix and GMix were set to 1 and 0.5, respectively. We use r = 0.5 for HMix. ... We train Res Net50 [26] with various MSDA methods for 300 epochs using SGD optimizer with a learning rate 0.1. We set the hyper-parameter α for all methods except Mixup to 1, while Mixup has α = 0.8. We use r = 0.75 for HMix.