Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Predicting the Next Action by Modeling the Abstract Goal

Debaditya Roy, Basura Fernando

TL;DR

This work tackles action anticipation by leveraging abstract goals inferred from visual sequences to reduce uncertainty in predicting the next action. It introduces a stochastic framework built on a variational recurrent network that learns a feature-based abstract goal from observed features and an action-based abstract goal from potential next actions, complemented by a goal-consistency criterion to select the most plausible next action. The approach achieves state-of-the-art results on EPIC-KITCHENS55 and EGTEA Gaze+ and demonstrates strong generalization to unseen kitchens, with ablations highlighting the importance of the GC loss and criterion. While effective, the method notes limitations in interpreting the learned abstract goals and struggles with long-tailed distributions in EK100, suggesting avenues for future work in interpretability and tail handling.

Abstract

The problem of anticipating human actions is an inherently uncertain one. However, we can reduce this uncertainty if we have a sense of the goal that the actor is trying to achieve. Here, we present an action anticipation model that leverages goal information for the purpose of reducing the uncertainty in future predictions. Since we do not possess goal information or the observed actions during inference, we resort to visual representation to encapsulate information about both actions and goals. Through this, we derive a novel concept called abstract goal which is conditioned on observed sequences of visual features for action anticipation. We design the abstract goal as a distribution whose parameters are estimated using a variational recurrent network. We sample multiple candidates for the next action and introduce a goal consistency measure to determine the best candidate that follows from the abstract goal. Our method obtains impressive results on the very challenging Epic-Kitchens55 (EK55), EK100, and EGTEA Gaze+ datasets. We obtain absolute improvements of +13.69, +11.24, and +5.19 for Top-1 verb, Top-1 noun, and Top-1 action anticipation accuracy respectively over prior state-of-the-art methods for seen kitchens (S1) of EK55. Similarly, we also obtain significant improvements in the unseen kitchens (S2) set for Top-1 verb (+10.75), noun (+5.84) and action (+2.87) anticipation. Similar trend is observed for EGTEA Gaze+ dataset, where absolute improvement of +9.9, +13.1 and +6.8 is obtained for noun, verb, and action anticipation. It is through the submission of this paper that our method is currently the new state-of-the-art for action anticipation in EK55 and EGTEA Gaze+ https://competitions.codalab.org/competitions/20071#results Code available at https://github.com/debadityaroy/Abstract_Goal

Predicting the Next Action by Modeling the Abstract Goal

TL;DR

This work tackles action anticipation by leveraging abstract goals inferred from visual sequences to reduce uncertainty in predicting the next action. It introduces a stochastic framework built on a variational recurrent network that learns a feature-based abstract goal from observed features and an action-based abstract goal from potential next actions, complemented by a goal-consistency criterion to select the most plausible next action. The approach achieves state-of-the-art results on EPIC-KITCHENS55 and EGTEA Gaze+ and demonstrates strong generalization to unseen kitchens, with ablations highlighting the importance of the GC loss and criterion. While effective, the method notes limitations in interpreting the learned abstract goals and struggles with long-tailed distributions in EK100, suggesting avenues for future work in interpretability and tail handling.

Abstract

The problem of anticipating human actions is an inherently uncertain one. However, we can reduce this uncertainty if we have a sense of the goal that the actor is trying to achieve. Here, we present an action anticipation model that leverages goal information for the purpose of reducing the uncertainty in future predictions. Since we do not possess goal information or the observed actions during inference, we resort to visual representation to encapsulate information about both actions and goals. Through this, we derive a novel concept called abstract goal which is conditioned on observed sequences of visual features for action anticipation. We design the abstract goal as a distribution whose parameters are estimated using a variational recurrent network. We sample multiple candidates for the next action and introduce a goal consistency measure to determine the best candidate that follows from the abstract goal. Our method obtains impressive results on the very challenging Epic-Kitchens55 (EK55), EK100, and EGTEA Gaze+ datasets. We obtain absolute improvements of +13.69, +11.24, and +5.19 for Top-1 verb, Top-1 noun, and Top-1 action anticipation accuracy respectively over prior state-of-the-art methods for seen kitchens (S1) of EK55. Similarly, we also obtain significant improvements in the unseen kitchens (S2) set for Top-1 verb (+10.75), noun (+5.84) and action (+2.87) anticipation. Similar trend is observed for EGTEA Gaze+ dataset, where absolute improvement of +9.9, +13.1 and +6.8 is obtained for noun, verb, and action anticipation. It is through the submission of this paper that our method is currently the new state-of-the-art for action anticipation in EK55 and EGTEA Gaze+ https://competitions.codalab.org/competitions/20071#results Code available at https://github.com/debadityaroy/Abstract_Goal
Paper Structure (20 sections, 19 equations, 1 figure, 8 tables, 1 algorithm)

This paper contains 20 sections, 19 equations, 1 figure, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Goals in action anticipation
  4. Features for action anticipation
  5. Past-Future Correlation for anticipation
  6. Action anticipation with abstract goals
  7. Feature-based abstract goal
  8. Action representations
  9. Action-based abstract goal representation
  10. Next action anticipation with goal consistency
  11. Loss functions and training of our model
  12. Experiments and results
  13. Datasets, features, and training details
  14. Comparison with state-of-the-art
  15. Impact of goal consistency criterion and loss
  16. ...and 5 more sections

Figures (1)

  • Figure 1: Illustration of the model design for abstract goal-based action anticipation. Yellow ellipses represent distributions and pink boxes represent various variables of the model. Best viewed on a screen.