Summary of Convolutional Two-stream Network Fusion For Video Action Recognition, by Christoph Feichtenhofer et al.
Convolutional Two-Stream Network Fusion for Video Action Recognition
by Christoph Feichtenhofer, Axel Pinz, Andrew Zisserman
First submitted to arxiv on: 22 Apr 2016
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed research explores various methods to combine the strengths of Convolutional Neural Networks (ConvNets) in processing spatial and temporal information for human action recognition in videos. By fusing ConvNet towers both spatially and temporally, the study reveals three key findings: Firstly, fusing at the convolution layer without loss of performance can save a substantial number of parameters. Secondly, spatial fusion at the last convolutional layer is more effective than earlier, with additional class prediction layer fusion boosting accuracy. Lastly, pooling abstract convolutional features over spatiotemporal neighbourhoods further enhances performance. Building on these discoveries, the authors propose a novel ConvNet architecture for spatiotemporal video snippet fusion and demonstrate state-of-the-art results on standard benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper investigates how to combine different parts of ConvNets to recognize human actions in videos. It shows that combining information from different parts can improve accuracy without increasing the number of parameters needed. The study also finds that combining information at certain points is better than others, and that adding an extra step can further improve results. Based on these findings, the researchers propose a new way to combine ConvNets for video action recognition. |
Keywords
» Artificial intelligence » Boosting » Spatiotemporal
