What have we learned from deep representations for action recognition?
As the success of deep models has led to their deployment in all areas of computer vision, it is increasingly important to understand how these representations work and what they are capturing. In this paper, we shed light on deep spatiotemporal representations by visualizing what two-stream models...
Main Authors: | , , , |
---|---|
Format: | Conference item |
Published: |
Institute of Electrical and Electronics Engineers
2018
|
_version_ | 1797067638158917632 |
---|---|
author | Feichtenhofer, C Pinz, A Wildes, R Zisserman, A |
author_facet | Feichtenhofer, C Pinz, A Wildes, R Zisserman, A |
author_sort | Feichtenhofer, C |
collection | OXFORD |
description | As the success of deep models has led to their deployment in all areas of computer vision, it is increasingly important to understand how these representations work and what they are capturing. In this paper, we shed light on deep spatiotemporal representations by visualizing what two-stream models have learned in order to recognize actions in video. We show that local detectors for appearance and motion objects arise to form distributed representations for recognizing human actions. Key observations include the following. First, cross-stream fusion enables the learning of true spatiotemporal features rather than simply separate appearance and motion features. Second, the networks can learn local representations that are highly class specific, but also generic representations that can serve a range of classes. Third, throughout the hierarchy of the network, features become more abstract and show increasing invariance to aspects of the data that are unimportant to desired distinctions (e.g. motion patterns across various speeds). Fourth, visualizations can be used not only to shed light on learned representations, but also to reveal idiosyncracies of training data and to explain failure cases of the system. |
first_indexed | 2024-03-06T21:59:12Z |
format | Conference item |
id | oxford-uuid:4e03a2a0-0124-4cde-bb99-a77ee88664b2 |
institution | University of Oxford |
last_indexed | 2024-03-06T21:59:12Z |
publishDate | 2018 |
publisher | Institute of Electrical and Electronics Engineers |
record_format | dspace |
spelling | oxford-uuid:4e03a2a0-0124-4cde-bb99-a77ee88664b22022-03-26T15:58:41ZWhat have we learned from deep representations for action recognition?Conference itemhttp://purl.org/coar/resource_type/c_5794uuid:4e03a2a0-0124-4cde-bb99-a77ee88664b2Symplectic Elements at OxfordInstitute of Electrical and Electronics Engineers2018Feichtenhofer, CPinz, AWildes, RZisserman, AAs the success of deep models has led to their deployment in all areas of computer vision, it is increasingly important to understand how these representations work and what they are capturing. In this paper, we shed light on deep spatiotemporal representations by visualizing what two-stream models have learned in order to recognize actions in video. We show that local detectors for appearance and motion objects arise to form distributed representations for recognizing human actions. Key observations include the following. First, cross-stream fusion enables the learning of true spatiotemporal features rather than simply separate appearance and motion features. Second, the networks can learn local representations that are highly class specific, but also generic representations that can serve a range of classes. Third, throughout the hierarchy of the network, features become more abstract and show increasing invariance to aspects of the data that are unimportant to desired distinctions (e.g. motion patterns across various speeds). Fourth, visualizations can be used not only to shed light on learned representations, but also to reveal idiosyncracies of training data and to explain failure cases of the system. |
spellingShingle | Feichtenhofer, C Pinz, A Wildes, R Zisserman, A What have we learned from deep representations for action recognition? |
title | What have we learned from deep representations for action recognition? |
title_full | What have we learned from deep representations for action recognition? |
title_fullStr | What have we learned from deep representations for action recognition? |
title_full_unstemmed | What have we learned from deep representations for action recognition? |
title_short | What have we learned from deep representations for action recognition? |
title_sort | what have we learned from deep representations for action recognition |
work_keys_str_mv | AT feichtenhoferc whathavewelearnedfromdeeprepresentationsforactionrecognition AT pinza whathavewelearnedfromdeeprepresentationsforactionrecognition AT wildesr whathavewelearnedfromdeeprepresentationsforactionrecognition AT zissermana whathavewelearnedfromdeeprepresentationsforactionrecognition |