Accepted author manuscript, 886 KB, PDF document
Accepted author manuscript
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
}
TY - GEN
T1 - ERA
T2 - Expert Retrieval and Assembly for Early Action Prediction
AU - Rahmani, Hossein
PY - 2022/10/27
Y1 - 2022/10/27
N2 - Early action prediction aims to successfully predict the classlabel of an action before it is completely performed. This is a challengingtask because the beginning stages of different actions can be very similar,with only minor subtle differences for discrimination. In this paper,we propose a novel Expert Retrieval and Assembly (ERA) module thatretrieves and assembles a set of experts most specialized at using discriminativesubtle differences, to distinguish an input sample from otherhighly similar samples. To encourage our model to effectively use subtledifferences for early action prediction, we push experts to discriminateexclusively between samples that are highly similar, forcing these expertsto learn to use subtle differences that exist between those samples.Additionally, we design an effective Expert Learning Rate Optimizationmethod that balances the experts’ optimization and leads to better performance.We evaluate our ERA module on four public action datasetsand achieve state-of-the-art performance.
AB - Early action prediction aims to successfully predict the classlabel of an action before it is completely performed. This is a challengingtask because the beginning stages of different actions can be very similar,with only minor subtle differences for discrimination. In this paper,we propose a novel Expert Retrieval and Assembly (ERA) module thatretrieves and assembles a set of experts most specialized at using discriminativesubtle differences, to distinguish an input sample from otherhighly similar samples. To encourage our model to effectively use subtledifferences for early action prediction, we push experts to discriminateexclusively between samples that are highly similar, forcing these expertsto learn to use subtle differences that exist between those samples.Additionally, we design an effective Expert Learning Rate Optimizationmethod that balances the experts’ optimization and leads to better performance.We evaluate our ERA module on four public action datasetsand achieve state-of-the-art performance.
M3 - Conference contribution/Paper
BT - European Conference on Computer Vision (ECCV)
PB - Springer
ER -