Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Ziyu, Xu, Yue, Lu, Cewu, Li, Yong-Lu
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914754841280512
author Wang, Ziyu
Xu, Yue
Lu, Cewu
Li, Yong-Lu
author_facet Wang, Ziyu
Xu, Yue
Lu, Cewu
Li, Yong-Lu
contents Recently, dataset distillation has paved the way towards efficient machine learning, especially for image datasets. However, the distillation for videos, characterized by an exclusive temporal dimension, remains an underexplored domain. In this work, we provide the first systematic study of video distillation and introduce a taxonomy to categorize temporal compression. Our investigation reveals that the temporal information is usually not well learned during distillation, and the temporal dimension of synthetic data contributes little. The observations motivate our unified framework of disentangling the dynamic and static information in the videos. It first distills the videos into still images as static memory and then compensates the dynamic and motion information with a learnable dynamic memory block. Our method achieves state-of-the-art on video datasets at different scales, with a notably smaller memory storage budget. Our code is available at https://github.com/yuz1wan/video_distillation.
format Preprint
id arxiv_https___arxiv_org_abs_2312_00362
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement
Wang, Ziyu
Xu, Yue
Lu, Cewu
Li, Yong-Lu
Computer Vision and Pattern Recognition
Machine Learning
Recently, dataset distillation has paved the way towards efficient machine learning, especially for image datasets. However, the distillation for videos, characterized by an exclusive temporal dimension, remains an underexplored domain. In this work, we provide the first systematic study of video distillation and introduce a taxonomy to categorize temporal compression. Our investigation reveals that the temporal information is usually not well learned during distillation, and the temporal dimension of synthetic data contributes little. The observations motivate our unified framework of disentangling the dynamic and static information in the videos. It first distills the videos into still images as static memory and then compensates the dynamic and motion information with a learnable dynamic memory block. Our method achieves state-of-the-art on video datasets at different scales, with a notably smaller memory storage budget. Our code is available at https://github.com/yuz1wan/video_distillation.
title Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2312.00362