{ "access": { "embargo": { "active": false, "reason": null }, "files": "public", "record": "public", "status": "open" }, "created": "2021-06-23T13:47:52.731395+00:00", "custom_fields": { "meeting:meeting": { "acronym": "AAAI'21", "title": "AAAI Conference on Artificial Intelligence" } }, "deletion_status": { "is_deleted": false, "status": "P" }, "files": { "count": 1, "enabled": true, "entries": { "AAAI-578.YangF.pdf": { "checksum": "md5:213460fc3930aa5bd848ebd74f0083fb", "ext": "pdf", "id": "a864b272-8bb1-47a9-b14a-bbd6cff42da4", "key": "AAAI-578.YangF.pdf", "metadata": null, "mimetype": "application/pdf", "size": 4604453 } }, "order": [], "total_bytes": 4604453 }, "id": "5018218", "is_draft": false, "is_published": true, "links": { "access": "https://zenodo.org/api/records/5018218/access", "access_links": "https://zenodo.org/api/records/5018218/access/links", "access_request": "https://zenodo.org/api/records/5018218/access/request", "access_users": "https://zenodo.org/api/records/5018218/access/users", "archive": "https://zenodo.org/api/records/5018218/files-archive", "archive_media": "https://zenodo.org/api/records/5018218/media-files-archive", "communities": "https://zenodo.org/api/records/5018218/communities", "communities-suggestions": "https://zenodo.org/api/records/5018218/communities-suggestions", "doi": "https://doi.org/10.5281/zenodo.5018218", "draft": "https://zenodo.org/api/records/5018218/draft", "files": "https://zenodo.org/api/records/5018218/files", "latest": "https://zenodo.org/api/records/5018218/versions/latest", "latest_html": "https://zenodo.org/records/5018218/latest", "media_files": "https://zenodo.org/api/records/5018218/media-files", "parent": "https://zenodo.org/api/records/5018217", "parent_doi": "https://zenodo.org/doi/10.5281/zenodo.5018217", "parent_html": "https://zenodo.org/records/5018217", "requests": "https://zenodo.org/api/records/5018218/requests", "reserve_doi": "https://zenodo.org/api/records/5018218/draft/pids/doi", "self": "https://zenodo.org/api/records/5018218", "self_doi": "https://zenodo.org/doi/10.5281/zenodo.5018218", "self_html": "https://zenodo.org/records/5018218", "self_iiif_manifest": "https://zenodo.org/api/iiif/record:5018218/manifest", "self_iiif_sequence": "https://zenodo.org/api/iiif/record:5018218/sequence/default", "versions": "https://zenodo.org/api/records/5018218/versions" }, "media_files": { "count": 0, "enabled": false, "entries": {}, "order": [], "total_bytes": 0 }, "metadata": { "creators": [ { "affiliations": [ { "name": "Xiamen University, China" } ], "person_or_org": { "family_name": "Fengxiang Yang", "name": "Fengxiang Yang", "type": "personal" } }, { "affiliations": [ { "name": "University of Trento, Italy" } ], "person_or_org": { "family_name": "Zhun Zhong", "name": "Zhun Zhong", "type": "personal" } }, { "affiliations": [ { "name": "National Institute of Informatics, Japan" } ], "person_or_org": { "family_name": "Hong Liu", "name": "Hong Liu", "type": "personal" } }, { "affiliations": [ { "name": "The University of Tokyo, Japan" } ], "person_or_org": { "family_name": "Zheng Wang", "name": "Zheng Wang", "type": "personal" } }, { "affiliations": [ { "name": "Xiamen University, China" } ], "person_or_org": { "family_name": "Zhiming Luo", "name": "Zhiming Luo", "type": "personal" } }, { "affiliations": [ { "name": "Xiamen University, China" } ], "person_or_org": { "family_name": "Shaozi Li", "name": "Shaozi Li", "type": "personal" } }, { "affiliations": [ { "name": "University of Trento, Italy" } ], "person_or_org": { "family_name": "Nicu Sebe", "name": "Nicu Sebe", "type": "personal" } }, { "affiliations": [ { "name": "National Institute of Informatics, Japan" } ], "person_or_org": { "family_name": "Shin'ichi Satoh", "name": "Shin'ichi Satoh", "type": "personal" } } ], "description": "
Recent advances in person re-identification (re-ID) have led to impressive retrieval accuracy. However, existing re-ID models are challenged by the adversarial examples crafted by adding quasi-imperceptible perturbations. Moreover, re-ID systems face the domain shift issue that training and testing domains are not consistent. In this study, we argue that learning powerful attackers with high universality that works well on unseen domains is an important step in promoting the robustness of re-ID systems. Therefore, we introduce a novel universal attack algorithm called “MetaAttack” for person re-ID. MetaAttack can mislead re-ID models on unseen domains by a universal adversarial perturbation. Specifically, to capture common patterns across different domains, we propose a meta-learning scheme to seek the universal perturbation via the gradient interaction between meta-train and meta-test formed by two datasets. We also take advantage of a virtual dataset (PersonX), instead of real ones, to conduct meta-test. This scheme not only enables us to learn with more comprehensive variation factors but also mitigates the negative effects caused by biased factors of real datasets. Experiments on three large-scale re-ID datasets demonstrate the effectiveness of our method in attacking re-ID models on unseen domains. Our final visualization results reveal some new properties of existing re-ID systems, which can guide us in designing a more robust re-ID model. Code and supplemental material are available at https://github.com/FlyingRoastDuck/MetaAttack AAAI21.
", "funding": [ { "award": { "acronym": "AI4Media", "id": "00k4n6c32::951911", "identifiers": [ { "identifier": "https://cordis.europa.eu/projects/951911", "scheme": "url" } ], "number": "951911", "program": "H2020", "title": { "en": "A European Excellence Centre for Media, Society and Democracy" } }, "funder": { "id": "00k4n6c32", "name": "European Commission" } } ], "publication_date": "2021-06-23", "publisher": "Zenodo", "resource_type": { "id": "publication-conferencepaper", "title": { "de": "Konferenzbeitrag", "en": "Conference paper" } }, "rights": [ { "description": { "en": "The Creative Commons Attribution license allows re-distribution and re-use of a licensed work on the condition that the creator is appropriately credited." }, "icon": "cc-by-icon", "id": "cc-by-4.0", "props": { "scheme": "spdx", "url": "https://creativecommons.org/licenses/by/4.0/legalcode" }, "title": { "en": "Creative Commons Attribution 4.0 International" } } ], "title": "Learning to Attack Real-World Models for Person Re-identification via Virtual-Guided Meta-Learning" }, "parent": { "access": { "owned_by": { "user": 237513 } }, "communities": { "default": "8538384b-cd84-4cc6-a77c-c259e9568dfd", "entries": [ { "access": { "member_policy": "open", "members_visibility": "public", "record_policy": "open", "review_policy": "open", "visibility": "public" }, "children": { "allow": false }, "created": "2020-10-22T12:35:38.515614+00:00", "custom_fields": {}, "deletion_status": { "is_deleted": false, "status": "P" }, "id": "8538384b-cd84-4cc6-a77c-c259e9568dfd", "links": {}, "metadata": { "curation_policy": "This community will accept all uploads that have been generated by partners of the AI4Media consortium that received funding from the project.
\r\n", "description": "Horizon H2020 EC-funded project AI4Media: A European Excellence Centre for Media, Society and Democracy", "page": "Motivated by the challenges, risks and opportunities that the widespread use of artificial intelligence (AI) has brought to the media, society and politics, the EU-funded AI4Media project aspires to establish a centre of excellence and a wide network of researchers across Europe and beyond. Its focus will be on delivering the next generation of core AI advances to serve the key sector of media, making sure that European values surrounding ethical and trustworthy AI are embedded in future AI deployments. The project will be supplemented by a funding framework, a PhD programme and a set of use cases to demonstrate the impact of the actions taken on the media sector.
\r\n", "title": "AI4Media H2020 Project" }, "revision_id": 0, "slug": "ai4media", "updated": "2020-10-22T12:35:38.584014+00:00" } ], "ids": [ "8538384b-cd84-4cc6-a77c-c259e9568dfd" ] }, "id": "5018217", "pids": { "doi": { "client": "datacite", "identifier": "10.5281/zenodo.5018217", "provider": "datacite" } } }, "pids": { "doi": { "client": "datacite", "identifier": "10.5281/zenodo.5018218", "provider": "datacite" }, "oai": { "identifier": "oai:zenodo.org:5018218", "provider": "oai" } }, "revision_id": 3, "stats": { "all_versions": { "data_volume": 336125069.0, "downloads": 73, "unique_downloads": 70, "unique_views": 88, "views": 105 }, "this_version": { "data_volume": 336125069.0, "downloads": 73, "unique_downloads": 70, "unique_views": 88, "views": 105 } }, "status": "published", "updated": "2021-06-24T01:48:21.147277+00:00", "versions": { "index": 1, "is_latest": true } }