Representation Flow for Action Recognition

AJ Piergiovanni, Michael S. Ryoo
10/2/2018
cs.CV

Official code: piergiaj/representation-flow-cvpr19 (254 stars, last commit 2022-09-12).

Abstract

In this paper, we propose a convolutional layer inspired by optical flow algorithms to learn motion representations. Our representation flow layer is a fully-differentiable layer designed to capture the `flow' of any representation channel within a convolutional neural network for action recognition. Its parameters for iterative flow optimization are learned in an end-to-end fashion together with the other CNN model parameters, maximizing the action recognition performance. Furthermore, we newly introduce the concept of learning `flow of flow' representations by stacking multiple representation flow layers. We conducted extensive experimental evaluations, confirming its advantages over previous recognition models using traditional optical flows in both computational speed and performance. Code/models available here: https://piergiaj.github.io/rep-flow-site/

View on arXivView PDF

Code Implementations(1)

piergiaj/representation-flow-cvpr19Official100%

Code and models for our CVPR'19 paper "Representation Flow for Action Recognition"

25459PythonFeb 27, 20194 years ago

Similar Papers

Flow Dynamics Correction for Action Recognition

87% match2023

ReynoldsFlow: Exquisite Flow Estimation via Reynolds Transport Theorem

87% match1 repo42025

GAFlow: Incorporating Gaussian Attention into Optical Flow

87% match1 repo422023

An End-to-End Two-Stream Network Based on RGB Flow and Representation Flow for Human Action Recognition

86% match2024

Flow Snapshot Neurons in Action: Deep Neural Networks Generalize to Biological Motion Perception

85% match1 repo2024

ScaleFlow++: Robust and Accurate Estimation of 3D Motion from Video

84% match1 repo412024

Cite this paper

@article{piergiovanni2018representation,
  title  = {Representation Flow for Action Recognition},
  author = {AJ Piergiovanni and Michael S. Ryoo},
  year   = {2018},
  eprint = {1810.01455},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1810.01455}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1810.01455.svg)](https://codeofpaper.com/paper/1810.01455)

Discussion