Last week was the time for ICCV 2017 in Venice, Italy. Setting aside my personal indifference to Venice (overcrowded and overpriced, uninspiring food), ICCV was an interesting meeting. It took place on Lido, at the same venue as the Venice Film Festival. It was certainly a more appealing location to me personally, but a bit of a way from San Marco and the tourist Venice.
The conference itself was, well, a mixed bag. Plenty of deep block shuffling works. About 3000 attendees is the number I heard.
We presented our work “PUnDA: Probabilistic Unsupervised Domain Adaptation for Knowledge Transfer Across Visual Categories” [1]. It shows that it is not always necessary to have end-to-end learning if the transfer model is properly constructed. As the luck has it, I had to reshuffle the schedule so the poster was both on Wednesday and Thursday (thanks to Robert Walecki, who help with the poster.) You can find a copy of the poster below.
[Bibtex]
@inproceedings{behnam17iccv,
added-at = {2017-07-18T17:26:03.000+0200},
author = {Gholami, Behnam and Rudovic, Ognjen and Pavlovic, Vladimir},
biburl = {https://www.bibsonomy.org/bibtex/29fd2aed9b960e71bf2472b5d70079060/vpavlovic},
booktitle = {Proc. IEEE International Conference Computer Vision},
interhash = {4a8c0ac9cfbd7b02e2d86ef7b5a289a0},
intrahash = {9fd2aed9b960e71bf2472b5d70079060},
keywords = {domain_adaptation myown unsupervised},
owner = {vladimir},
timestamp = {2017-07-18T18:21:19.000+0200},
title = {PUnDA: Probabilistic Unsupervised Domain Adaptation},
year = 2017
}