Multi-Stage LSTM (MS-LSTM) for Action Anticipation

Tools Click here to view this collection in the new DAP user interface

show summary fields  |   show all    

About this Collection

Multi-Stage LSTM (MS-LSTM) for Action Anticipation

Multi-Stage LSTM (MS-LSTM) is a deep sequence learning architecture that leverage two types of descriptors, i.e., action-aware and context-aware features to determine the action label. To do it at early stages, MS-LSTM is trained with a novel loss function that encourages correct prediction as early as possible.

Computer Vision Knowledge Representation and Machine Learning

Mohammad Sadegh Aliakbarian

Action Anticipation Deep Learning Long Short-Term Memory Computer Vision


Encouraging LSTMs To Anticipate Actions Very Early


Source code accompanying the publication of a conference paper at the 2017 International Conference on Computer Vision

Mohammad Sadegh Aliakbarian, Fatemeh Sadat Saleh, Mathieu Salzmann, Basura Fernando, Lars Petersson, Lars Andersson, 2017, Encouraging LSTMs To Anticipate Actions Very Early. In Proc. 2017 Int. Conf. Comput. Vision (ICCV), IEEE

GPLv3 Licence with CSIRO Disclaimer

CSIRO (Australia), Australian Centre For Robotic Vision (Australia), Australian National University (Australia)

Aliakbarian, Mohammad Sadegh; Saleh, Fatemeh Sadat; Salzmann, Mathieu; Fernando, Basura; Petersson, Lars; Andersson, Lars (2018): Multi-Stage LSTM (MS-LSTM) for Action Anticipation. v1. CSIRO. Software Collection.

All Rights (including copyright) CSIRO 2018.

The metadata and files (if any) are available to the public.

show all

About this Project


Mohammad Sadegh Aliakbarian

Fatemeh Sadat Saleh

Mathieu Salzmann

Basura Fernando

Lars Petersson

Lars Andersson

Others were also interested in

  • Synthetic Data for Non-rigid 3D Reconstruction using a Moving RGB-D Camera....
  • The Globally-Optimal Pose And Correspondences (GOPAC) Algorithm....
  • Data61 Pedestrian Dataset....