Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling
This paper describes an unsupervised sequential auto-encoding model targeting multi-object scenes. The proposed model uses an attention-based formulation, with reconstruction-driven losses. The main model relies on iteratively writing regions onto a canvas, in a differentiable manner. To enforce att...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Gazi University
2022-12-01
|
| Series: | Gazi Üniversitesi Fen Bilimleri Dergisi |
| Subjects: | |
| Online Access: | https://dergipark.org.tr/tr/pub/gujsc/issue/74502/1139701 |
| _version_ | 1797372253240819712 |
|---|---|
| author | Yarkın Deniz ÇETİN Ramazan Gökberk CİNBİŞ |
| author_facet | Yarkın Deniz ÇETİN Ramazan Gökberk CİNBİŞ |
| author_sort | Yarkın Deniz ÇETİN |
| collection | DOAJ |
| description | This paper describes an unsupervised sequential auto-encoding model targeting multi-object scenes. The proposed model uses an attention-based formulation, with reconstruction-driven losses. The main model relies on iteratively writing regions onto a canvas, in a differentiable manner. To enforce attention to objects and/or parts, the model uses a convolutional localization network, a region level bottleneck auto-encoder and a loss term that encourages reconstruction within a limited number of iterations. An extended version of the model incorporates a background modeling component that aims at handling scenes with complex backgrounds. The model is evaluated on two separate datasets: a synthetic dataset that is constructed by composing MNIST digit instances together, and the MS-COCO dataset. The model achieves high reconstruction ability on MNIST based scenes. The extended model shows promising results on the complex and challenging MS-COCO scenes. |
| first_indexed | 2024-03-08T18:33:07Z |
| format | Article |
| id | doaj.art-d42316933329442b88b81fae268f3025 |
| institution | Directory Open Access Journal |
| issn | 2147-9526 |
| language | English |
| last_indexed | 2024-03-08T18:33:07Z |
| publishDate | 2022-12-01 |
| publisher | Gazi University |
| record_format | Article |
| series | Gazi Üniversitesi Fen Bilimleri Dergisi |
| spelling | doaj.art-d42316933329442b88b81fae268f30252023-12-29T21:56:39ZengGazi UniversityGazi Üniversitesi Fen Bilimleri Dergisi2147-95262022-12-011041127114210.29109/gujsc.1139701 Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene ModelingYarkın Deniz ÇETİN0https://orcid.org/0000-0003-1358-4247Ramazan Gökberk CİNBİŞ1https://orcid.org/0000-0003-0962-7101İHSAN DOĞRAMACI BİLKENT ÜNİVERSİTESİORTA DOĞU TEKNİK ÜNİVERSİTESİThis paper describes an unsupervised sequential auto-encoding model targeting multi-object scenes. The proposed model uses an attention-based formulation, with reconstruction-driven losses. The main model relies on iteratively writing regions onto a canvas, in a differentiable manner. To enforce attention to objects and/or parts, the model uses a convolutional localization network, a region level bottleneck auto-encoder and a loss term that encourages reconstruction within a limited number of iterations. An extended version of the model incorporates a background modeling component that aims at handling scenes with complex backgrounds. The model is evaluated on two separate datasets: a synthetic dataset that is constructed by composing MNIST digit instances together, and the MS-COCO dataset. The model achieves high reconstruction ability on MNIST based scenes. The extended model shows promising results on the complex and challenging MS-COCO scenes.https://dergipark.org.tr/tr/pub/gujsc/issue/74502/1139701unsupervised learningcomplex scene modelingobject discovery |
| spellingShingle | Yarkın Deniz ÇETİN Ramazan Gökberk CİNBİŞ Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling Gazi Üniversitesi Fen Bilimleri Dergisi unsupervised learning complex scene modeling object discovery |
| title | Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling |
| title_full | Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling |
| title_fullStr | Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling |
| title_full_unstemmed | Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling |
| title_short | Attentive Sequential Auto-Encoding Towards Unsupervised Object-centric Scene Modeling |
| title_sort | attentive sequential auto encoding towards unsupervised object centric scene modeling |
| topic | unsupervised learning complex scene modeling object discovery |
| url | https://dergipark.org.tr/tr/pub/gujsc/issue/74502/1139701 |
| work_keys_str_mv | AT yarkındenizcetin attentivesequentialautoencodingtowardsunsupervisedobjectcentricscenemodeling AT ramazangokberkcinbis attentivesequentialautoencodingtowardsunsupervisedobjectcentricscenemodeling |