Finding an unsupervised image segmenter in each of your deep generative models
Recent research has shown that numerous human-interpretable directions exist in the latent space of GANs. In this paper, we develop an automatic procedure for finding directions that lead to foreground-background image separation, and we use these directions to train an image segmentation model with...
| Հիմնական հեղինակներ: | , , , |
|---|---|
| Ձևաչափ: | Conference item |
| Լեզու: | English |
| Հրապարակվել է: |
OpenReview
2021
|
| _version_ | 1826308567478566912 |
|---|---|
| author | Melas-Kyriazi, L Rupprecht, C Laina, I Vedaldi, A |
| author_facet | Melas-Kyriazi, L Rupprecht, C Laina, I Vedaldi, A |
| author_sort | Melas-Kyriazi, L |
| collection | OXFORD |
| description | Recent research has shown that numerous human-interpretable directions exist in the latent space of GANs. In this paper, we develop an automatic procedure for finding directions that lead to foreground-background image separation, and we use these directions to train an image segmentation model without human supervision. Our method is generator-agnostic, producing strong segmentation results with a wide range of different GAN architectures. Furthermore, by leveraging GANs pretrained on large datasets such as ImageNet, we are able to segment images from a range of domains without further training or finetuning. Evaluating our method on image segmentation benchmarks, we compare favorably to prior work while using neither human supervision nor access to the training data. Broadly, our results demonstrate that automatically extracting foreground-background structure from pretrained deep generative models can serve as a remarkably effective substitute for human supervision. |
| first_indexed | 2024-03-07T07:21:22Z |
| format | Conference item |
| id | oxford-uuid:78283170-463a-40fa-ab3c-496f90378e96 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T07:21:22Z |
| publishDate | 2021 |
| publisher | OpenReview |
| record_format | dspace |
| spelling | oxford-uuid:78283170-463a-40fa-ab3c-496f90378e962022-10-10T10:49:27ZFinding an unsupervised image segmenter in each of your deep generative modelsConference itemhttp://purl.org/coar/resource_type/c_6670uuid:78283170-463a-40fa-ab3c-496f90378e96EnglishSymplectic ElementsOpenReview2021Melas-Kyriazi, LRupprecht, CLaina, IVedaldi, ARecent research has shown that numerous human-interpretable directions exist in the latent space of GANs. In this paper, we develop an automatic procedure for finding directions that lead to foreground-background image separation, and we use these directions to train an image segmentation model without human supervision. Our method is generator-agnostic, producing strong segmentation results with a wide range of different GAN architectures. Furthermore, by leveraging GANs pretrained on large datasets such as ImageNet, we are able to segment images from a range of domains without further training or finetuning. Evaluating our method on image segmentation benchmarks, we compare favorably to prior work while using neither human supervision nor access to the training data. Broadly, our results demonstrate that automatically extracting foreground-background structure from pretrained deep generative models can serve as a remarkably effective substitute for human supervision. |
| spellingShingle | Melas-Kyriazi, L Rupprecht, C Laina, I Vedaldi, A Finding an unsupervised image segmenter in each of your deep generative models |
| title | Finding an unsupervised image segmenter in each of your deep generative models |
| title_full | Finding an unsupervised image segmenter in each of your deep generative models |
| title_fullStr | Finding an unsupervised image segmenter in each of your deep generative models |
| title_full_unstemmed | Finding an unsupervised image segmenter in each of your deep generative models |
| title_short | Finding an unsupervised image segmenter in each of your deep generative models |
| title_sort | finding an unsupervised image segmenter in each of your deep generative models |
| work_keys_str_mv | AT melaskyriazil findinganunsupervisedimagesegmenterineachofyourdeepgenerativemodels AT rupprechtc findinganunsupervisedimagesegmenterineachofyourdeepgenerativemodels AT lainai findinganunsupervisedimagesegmenterineachofyourdeepgenerativemodels AT vedaldia findinganunsupervisedimagesegmenterineachofyourdeepgenerativemodels |