Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.

Representation Flow for Action Recognition

This repository contains the code for our CVPR 2019 paper, Project Page:

AJ Piergiovanni and Michael S. Ryoo
"Representation Flow for Action Recognition"
in CVPR 2019

If you find the code useful for your research, please cite our paper:

          title={Representation Flow for Action Recognition},
          booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
          author={AJ Piergiovanni and Michael S. Ryoo},

Representation Flow Layer


We introduce the representation flow layer, which can be found in This layer iteratively estimates the flow, can be applied to CNN feature maps, and is fully learnable to maximize classification performance.

Activity Recognition Experiments

model overview

We include our pretrained models for HMDB and Kinetics in models/.

We tested our models on HMDB and Kinetics and provide the files and class labes used in data.


Method Kinetics-400 HMDB Runtime
2D Two-Stream 64.5 66.6 8546ms
TVNet (+RGB) - 71.0 785ms
(2+1)D Two-Stream 75.4 78.7 8623ms
I3D Two-stream 74.2 80.7 9354ms
(2+1)D + Rep-Flow 75.5 77.1 622ms
(2+1)D + Flow-of-flow 77.1 81.1 654ms

Visualization of learned flows

Examples of representation flows for various actions. The representation flow is computed after the 3rd residual block and captures some sematic motion information. At this point, the representations are low dimensional (28x28).

Examples of representation flows for different channels for "clapping." Some channels capture the hand motion, while other channels focus on different features/motion patterns not present in this clip.


Our code has been tested on Ubuntu 14.04 and 16.04 using python 3.6, PyTorch version 0.4.1 with a Titan X GPU. Our data loading uses lintel to extract frames from the video.


  1. Download the code git clone

  2. Pre-trained models are avialable here.

  3. contains the code to train and evaluate models.


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






No releases published


No packages published