abstracts[] |
{'sha1': '16f041cf048b10909256e968d7875b10a65e12a9', 'content': 'Automated photo tagging has established itself as one of the most compelling\napplications of deep learning. While deep convolutional neural networks have\nrepeatedly demonstrated top performance on standard datasets for\nclassification, there are a number of often overlooked but important\nconsiderations when deploying this technology in a real-world scenario. In this\npaper, we present our efforts in developing a large-scale photo tagging system\nfor Flickr photo search. We discuss topics including how to 1) select the tags\nthat matter most to our users; 2) develop lightweight, high-performance models\nfor tag prediction; and 3) leverage the power of large amounts of noisy data\nfor training. Our results demonstrate that, for real-world datasets, training\nexclusively with this noisy data yields performance on par with the standard\nparadigm of first pre-training on clean data and then fine-tuning. In addition,\nwe observe that the models trained with user-generated data can yield better\nfine-tuning results when a small amount of clean data is available. As such, we\nadvocate for the approach of harnessing user-generated data in large-scale\nsystems.', 'mimetype': 'text/plain', 'lang': 'en'}
|
container |
|
container_id |
|
contribs[] |
{'index': 0, 'creator_id': None, 'creator': None, 'raw_name': 'Kofi Boakye', 'given_name': None, 'surname': None, 'role': 'author', 'raw_affiliation': None, 'extra': None}
{'index': 1, 'creator_id': None, 'creator': None, 'raw_name': 'Sachin Farfade', 'given_name': None, 'surname': None, 'role': 'author', 'raw_affiliation': None, 'extra': None}
{'index': 2, 'creator_id': None, 'creator': None, 'raw_name': 'Hamid Izadinia', 'given_name': None, 'surname': None, 'role': 'author', 'raw_affiliation': None, 'extra': None}
{'index': 3, 'creator_id': None, 'creator': None, 'raw_name': 'Yannis Kalantidis', 'given_name': None, 'surname': None, 'role': 'author', 'raw_affiliation': None, 'extra': None}
{'index': 4, 'creator_id': None, 'creator': None, 'raw_name': 'and\n Pierre Garrigues', 'given_name': None, 'surname': None, 'role': 'author', 'raw_affiliation': None, 'extra': None}
|
ext_ids |
{'doi': None, 'wikidata_qid': None, 'isbn13': None, 'pmid': None, 'pmcid': None, 'core': None, 'arxiv': '1612.01922v3', 'jstor': None, 'ark': None, 'mag': None, 'doaj': None, 'dblp': None, 'oai': None, 'hdl': None}
|
files[] |
{'state': 'active', 'ident': 'bd7np2fupbd7nfp6amxjxznsvq', 'revision': '01fcfa38-cf83-40d3-bade-0155caa70987', 'redirect': None, 'extra': None, 'edit_extra': None, 'size': 9169564, 'md5': '6d6893d17bcfda0f33fc56ac300ade3c', 'sha1': '302d1368364192bbd5da90e96b376d06a98dac26', 'sha256': '434c485d86bc5b843fa588712c91bbe86c8e19889716e3958f2fb11fe97ebcc0', 'urls': [{'url': 'https://arxiv.org/pdf/1612.01922v3.pdf', 'rel': 'repository'}, {'url': 'https://web.archive.org/web/20200831164729/https://arxiv.org/pdf/1612.01922v3.pdf', 'rel': 'webarchive'}], 'mimetype': 'application/pdf', 'content_scope': None, 'release_ids': ['y5ek65nfancwpfg3ddk4txc7wq'], 'releases': None}
|
filesets |
[]
|
issue |
|
language |
en
|
license_slug |
ARXIV-1.0
|
number |
|
original_title |
|
pages |
|
publisher |
|
refs |
[]
|
release_date |
2017-12-19
|
release_stage |
submitted
|
release_type |
article
|
release_year |
2017
|
subtitle |
|
title |
Tag Prediction at Flickr: a View from the Darkroom
|
version |
v3
|
volume |
|
webcaptures |
[]
|
withdrawn_date |
|
withdrawn_status |
|
withdrawn_year |
|
work_id |
abdzx2wgurbv5aecztydisoifa
|