Learning Hierarchical Invariant Spatio-temporal Features for Action Recognition
- Login to Download
- 1 Credits
Resource Overview
Learning Hierarchical Invariant Spatio-temporal Features for Action Recognition using Independent Subspace Analysis. This package provides ready-to-use video features generated by the stacked convolutional ISA model described in the paper, along with functions for direct feature extraction from video data, including implementation of the multi-layer ISA architecture and feature vector computation algorithms.
Detailed Documentation
In this paper, we introduce an approach called Independent Subspace Analysis (ISA) for learning hierarchical invariant spatio-temporal features for action recognition. We provide a software package that delivers ready-to-use video features using the stacked convolutional ISA model described in our research, along with comprehensive functions for extracting features directly from video inputs. The package implements a multi-layer architecture where each level applies ISA to learn increasingly complex feature representations, with key functions handling video preprocessing, convolutional filtering, and feature aggregation across temporal windows. This toolkit significantly streamlines your workflow, enabling more efficient research and development in action recognition by providing optimized implementations of the feature extraction pipeline and spatial-temporal pattern learning algorithms.
- Login to Download
- 1 Credits