Thesis
Self-supervised learning using motion and visualizing convolutional neural networks
- Abstract:
-
We propose a novel method for learning convolutional image representations without manual supervision. We use motion in the form of optical-flow, to supervise representations of static images. Training a network to predict flow from a single image can be needlessly difficult due to intrinsic ambiguities in this prediction task. We instead propose two simpler learning goals: (a) embed pixels such that the similarity between their embeddings matches that between their optical-flow vectors (C...
Expand abstract
Actions
Access Document
- Files:
-
-
(pdf, 52.1MB)
-
Why is the content I wish to access not available via ORA?
Bibliographic data (the information relating to research outputs) and full-text items (e.g. articles, theses, reports, etc.) arrive in ORA from several different sources. Unfortunately we are not able to make available the full-text for every research output.
Please contact the ORA team (
) if you have queries regarding unavailable content OR if you are aware of a full-text copy we can make available.
) if you have queries regarding unavailable content OR if you are aware of a full-text copy we can make available.
Content may be unavailable for the following four reasons
-
Version unsuitable
-
We have not obtained a suitable full-text for a given research output. See this page for more information.
-
Recently completed
-
Sometimes content is held in ORA but is unavailable for a fixed period of time to comply with the policies and wishes of rights holders.
-
Permissions
-
All content made available in ORA should comply with relevant rights, such as copyright. See this page for more information.
-
Clearance
-
Some thesis volumes scanned as part of the digitisation scheme funded by Dr Leonard Polonsky are currently unavailable due to sensitive material or uncleared third-party copyright content. We are attempting to contact authors whose theses are affected.
Alternative access to the full-text
You may be able to access the full-text directly from the publisher's website using the 'Publisher Copy' link in the 'Links & Downloads' box from a research output's ORA record page. This method may require an institutional or individual subscription to the journal/resource.
Authors
Contributors
+ Vedaldi, A
Department:
University of Oxford
Role:
Supervisor
+ Zisserman, A
Department:
University of Oxford
Role:
Examiner
+ Favaro, P
Department:
Universität Bern
Role:
Examiner
Funding
+ European Research Council
More from this funder
Funding agency for:
Mahendran, A
Grant:
677195-IDIU
Bibliographic Details
- Type of award:
- DPhil
- Level of award:
- Doctoral
- Awarding institution:
- University of Oxford
Item Description
- Language:
- English
- Keywords:
- Subjects:
- UUID:
-
uuid:05ef7004-0bb1-4852-be1f-892daf694430
- Deposit date:
- 2019-02-23
Related Items
Terms of use
- Copyright holder:
- Mahendran, A
- Copyright date:
- 2018
If you are the owner of this record, you can report an update to it here: Report update to this record