search

unizard / AwesomeArxiv

89 stars 14 forks source link

[2017.10.11][***] Learning semantic relationships for better action retrieval in images #21

Open unizard opened 7 years ago

unizard commented 7 years ago

CVPR2015

Institute: Stanford, Google URL: paper, Supp Keywords: Interest: 3

unizard commented 7 years ago

Weakly-supervised learning of visual relations, action 이 아니다?

Vision - Language 의 관계를 보는 것이다.

unizard commented 7 years ago

Related Work

Devise: A Deep Visual-Semantic Embedding Model, NIPs2013

https://static.googleusercontent.com/media/research.google.com/ko//pubs/archive/41869.pdf http://vision.cs.utexas.edu/381V-fall2016/slides/nagarajan-paper.pdf http://www.cs.virginia.edu/~vicente/vislang/slides/devise.pdf http://videolectures.net/nipsworkshops2013_bengio_embedding_model/

https://github.com/gtlim/DeVise https://github.com/akshaychawla/devise-keras

unizard commented 7 years ago

관련연구 #인용수295 #코드공개해서그런듯

Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models, TACL 2015.

https://arxiv.org/abs/1411.2539 http://deeplearning.cs.toronto.edu/i2t https://github.com/ryankiros/visual-semantic-embedding

unizard commented 7 years ago

관련연구 #Best Paper Award

Large-scale object classification using label relation graphs, ECCV2014

http://web.eecs.umich.edu/~jiadeng/paper/deng2014large.pdf https://github.com/ronghanghu/hex_graph https://docs.google.com/presentation/d/1Gl60i1dyZ8iyrNbyPKUfeXkOL2pwe8i0hY8Bn4E-FNw/edit#slide=id.g47ac826c7_0179 https://www.slideshare.net/takmin/ieee2014

unizard commented 7 years ago

우리는 아주 오랫동안 영상 이해를 위해

대상과 대상의 관계(위치관계/상호작용여부/)를 대상의 장소를 대상의 특정 행동을 적극 활용해 왔다. 지금 어느 단계 까지 와 있을까? 나는 Multiframe을 통한 연구가 아닌, single image 기반의 이해를 원한다.

Captioning 연구 VQA 연구

Object-Object Relation 연구

Action-Action Relation

Human-Object Relation

unizard commented 7 years ago

Weakly supervised learning of visual relations, 2017/07/CoRR https://arxiv.org/pdf/1707.09472.pdf

unizard commented 7 years ago

Facebook3인방 #메이저낙방인가

[IntractNet]Detecting and Recognizing Human-Object Interactions, CoRR. Multi-task learning: Action classification + Localization

https://arxiv.org/abs/1704.07333 https://gkioxari.github.io/InteractNet/index.html https://www.facebook.com/groups/TensorFlowKR/permalink/462002037474193/

unizard commented 7 years ago

VisualRelationships

Detecting Visual Relationships with Deep Relational Networks, CVPR2017

https://arxiv.org/abs/1704.03114 https://www.youtube.com/watch?v=ffpy3BqN88o https://github.com/doubledaibo/drnet_cvpr2017

http://xuewenyuan.github.io/2017/07/29/Detecting-Visual-Relationships-with-Deep-Relational-Networks-%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0/

unizard commented 7 years ago

ICCV2017 #Interaction #code

Towards Context-aware Interaction Recognition for Visual Relationship Detection, ICCV2017 Care about you: towards large-scale human-centric visual relationship detection, CoRR2017

https://arxiv.org/pdf/1703.06246.pdf https://arxiv.org/pdf/1705.09892.pdf

https://bitbucket.org/jingruixiaozhuang/iccv2017_vrd/src

unizard commented 7 years ago

CVPR2017 #Naver-Labs-Europe

Beyond Instance-Level Image Retrieval: Leveraging Captions to Learn a Global Visual Representation for Semantic Retrieval, CVPR2017.

http://openaccess.thecvf.com/content_cvpr_2017/papers/Gordo_Beyond_Instance-Level_Image_CVPR_2017_paper.pdf

https://norman3.github.io/papers/docs/deepimgir <- 네이버 비전팀 홍기호님자료