ECCV 2024 Day4 - Oral: Dataset Condensation

 

 

 

ECCV 2024. 10. 02. Wednesday 
Oral 3A: Datasets And Benchmarking

 

 

오늘은 꼭 들어야 할 주제가 있기 때문에,

오전 발표를 듣기 위해, 일찍 나섰습니다 ☁️

 

 

 

☔️

 

 

 

입구에서 티셔츠 수령 👕

 

 

 

Oral: Dataset Condensation 

Towards Model-Agnostic Dataset Condensation by Heterogeneous Models

Jun-Yeong Moon · Jung Uk Kim · Gyeong-Moon Park

Gold Room

 

 

 

교수님들 얼굴이 뙇 보여서 먼가 신기했답,, :)

 

 

 

https://eccv.ecva.net/virtual/2024/oral/856

ECCV 2024 Towards Model-Agnostic Dataset Condensation by Heterogeneous Models Oral

 

 

 

오늘은 노트북 안챙겨가고, 모바일로 기록했다!

 

 

 

Abstract

 The advancement of deep learning has coincided with the proliferation of both models and available data. The surge in dataset sizes and the subsequent surge in computational requirements have led to the development of the Dataset Condensation (DC). While prior studies have delved into generating synthetic images through methods like distribution alignment and training trajectory tracking for more efficient model training, a significant challenge arises when employing these condensed images practically. Notably, these condensed images tend to be specific to particular models, constraining their versatility and practicality.

 In response to this limitation, we introduce a novel method, Heterogeneous Model Dataset Condensation (HMDC), designed to produce universally applicable condensed images through cross-model interactions. To address the issues of gradient magnitude difference and semantic distance in models when utilizing heterogeneous models, we propose the Gradient Balance Module (GBM) and Mutual Distillation (MD) with the Spatial-Semantic Decomposition method. By balancing the contribution of each model and maintaining their semantic meaning closely, our approach overcomes the limitations associated with model-specific condensed images and enhances the broader utility.

 

 

 

Introduction

 

 

 

Methods

 

 

 

여러 모델을 활용

 

 

 

 

 

 

기존 DC의 문제점을 두 모델을 모두 활용하여 개선한 방법

 

 

 

🔗 Git-hub Link

https://github.com/KHU-AGI/HMDC

GitHub - KHU-AGI/HMDC: Official Repository for Heterogeneous Models Dataset Condensation (ECCV 2024, Oral)