Dynamic graph message passing networks
Modelling long-range dependencies is critical for scene understanding tasks in computer vision. Although CNNs have excelled in many vision tasks, they are still limited in capturing long-range structured relationships as they typically consist of layers of local kernels. A fully-connected graph is b...
| Main Authors: | , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
IEEE
2020
|
| _version_ | 1826282152976711680 |
|---|---|
| author | Zhang, L Xu, D Arnab, A Torr, PHS |
| author_facet | Zhang, L Xu, D Arnab, A Torr, PHS |
| author_sort | Zhang, L |
| collection | OXFORD |
| description | Modelling long-range dependencies is critical for scene understanding tasks in computer vision. Although CNNs have excelled in many vision tasks, they are still limited in capturing long-range structured relationships as they typically consist of layers of local kernels. A fully-connected graph is beneficial for such modelling, however, its computational overhead is prohibitive. We propose a dynamic graph message passing network, that significantly reduces the computational complexity compared to related works modelling a fully-connected graph. This is achieved by adaptively sampling nodes in the graph, conditioned on the input, for message passing. Based on the sampled nodes, we dynamically predict node-dependent filter weights and the affinity matrix for propagating information between them. Using this model, we show significant improvements with respect to strong, state-of-the-art baselines on three different tasks and backbone architectures. Our approach also outperforms fully-connected graphs while using substantially fewer floating-point operations and parameters. |
| first_indexed | 2024-03-07T00:39:32Z |
| format | Conference item |
| id | oxford-uuid:828ecf0f-d1cf-46b5-bf31-3c7fe38b17db |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T00:39:32Z |
| publishDate | 2020 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:828ecf0f-d1cf-46b5-bf31-3c7fe38b17db2022-03-26T21:38:24ZDynamic graph message passing networksConference itemhttp://purl.org/coar/resource_type/c_5794uuid:828ecf0f-d1cf-46b5-bf31-3c7fe38b17dbEnglishSymplectic ElementsIEEE2020Zhang, LXu, DArnab, ATorr, PHSModelling long-range dependencies is critical for scene understanding tasks in computer vision. Although CNNs have excelled in many vision tasks, they are still limited in capturing long-range structured relationships as they typically consist of layers of local kernels. A fully-connected graph is beneficial for such modelling, however, its computational overhead is prohibitive. We propose a dynamic graph message passing network, that significantly reduces the computational complexity compared to related works modelling a fully-connected graph. This is achieved by adaptively sampling nodes in the graph, conditioned on the input, for message passing. Based on the sampled nodes, we dynamically predict node-dependent filter weights and the affinity matrix for propagating information between them. Using this model, we show significant improvements with respect to strong, state-of-the-art baselines on three different tasks and backbone architectures. Our approach also outperforms fully-connected graphs while using substantially fewer floating-point operations and parameters. |
| spellingShingle | Zhang, L Xu, D Arnab, A Torr, PHS Dynamic graph message passing networks |
| title | Dynamic graph message passing networks |
| title_full | Dynamic graph message passing networks |
| title_fullStr | Dynamic graph message passing networks |
| title_full_unstemmed | Dynamic graph message passing networks |
| title_short | Dynamic graph message passing networks |
| title_sort | dynamic graph message passing networks |
| work_keys_str_mv | AT zhangl dynamicgraphmessagepassingnetworks AT xud dynamicgraphmessagepassingnetworks AT arnaba dynamicgraphmessagepassingnetworks AT torrphs dynamicgraphmessagepassingnetworks |