Matches in SemOpenAlex for { <https://semopenalex.org/work/W4287243477> ?p ?o ?g. }
Showing items 1 to 72 of
72
with 100 items per page.
- W4287243477 abstract "Instance segmentation models today are very accurate when trained on large annotated datasets, but collecting mask annotations at scale is prohibitively expensive. We address the partially supervised instance segmentation problem in which one can train on (significantly cheaper) bounding boxes for all categories but use masks only for a subset of categories. In this work, we focus on a popular family of models which apply differentiable cropping to a feature map and predict a mask based on the resulting crop. Under this family, we study Mask R-CNN and discover that instead of its default strategy of training the mask-head with a combination of proposals and groundtruth boxes, training the mask-head with only groundtruth boxes dramatically improves its performance on novel classes. This training strategy also allows us to take advantage of alternative mask-head architectures, which we exploit by replacing the typical mask-head of 2-4 layers with significantly deeper off-the-shelf architectures (e.g. ResNet, Hourglass models). While many of these architectures perform similarly when trained in fully supervised mode, our main finding is that they can generalize to novel classes in dramatically different ways. We call this ability of mask-heads to generalize to unseen classes the strong mask generalization effect and show that without any specialty modules or losses, we can achieve state-of-the-art results in the partially supervised COCO instance segmentation benchmark. Finally, we demonstrate that our effect is general, holding across underlying detection methodologies (including anchor-based, anchor-free or no detector at all) and across different backbone networks. Code and pre-trained models are available at https://git.io/deepmac." @default.
- W4287243477 created "2022-07-25" @default.
- W4287243477 creator A5008019088 @default.
- W4287243477 creator A5027687139 @default.
- W4287243477 creator A5076825006 @default.
- W4287243477 creator A5079026336 @default.
- W4287243477 creator A5089510841 @default.
- W4287243477 date "2021-04-01" @default.
- W4287243477 modified "2023-10-06" @default.
- W4287243477 title "The surprising impact of mask-head architecture on novel class segmentation" @default.
- W4287243477 hasPublicationYear "2021" @default.
- W4287243477 type Work @default.
- W4287243477 citedByCount "0" @default.
- W4287243477 crossrefType "posted-content" @default.
- W4287243477 hasAuthorship W4287243477A5008019088 @default.
- W4287243477 hasAuthorship W4287243477A5027687139 @default.
- W4287243477 hasAuthorship W4287243477A5076825006 @default.
- W4287243477 hasAuthorship W4287243477A5079026336 @default.
- W4287243477 hasAuthorship W4287243477A5089510841 @default.
- W4287243477 hasBestOaLocation W42872434771 @default.
- W4287243477 hasConcept C119857082 @default.
- W4287243477 hasConcept C13280743 @default.
- W4287243477 hasConcept C134306372 @default.
- W4287243477 hasConcept C138885662 @default.
- W4287243477 hasConcept C153180895 @default.
- W4287243477 hasConcept C154945302 @default.
- W4287243477 hasConcept C165696696 @default.
- W4287243477 hasConcept C177148314 @default.
- W4287243477 hasConcept C185798385 @default.
- W4287243477 hasConcept C205649164 @default.
- W4287243477 hasConcept C2776401178 @default.
- W4287243477 hasConcept C2777212361 @default.
- W4287243477 hasConcept C33923547 @default.
- W4287243477 hasConcept C38652104 @default.
- W4287243477 hasConcept C41008148 @default.
- W4287243477 hasConcept C41895202 @default.
- W4287243477 hasConcept C63584917 @default.
- W4287243477 hasConcept C89600930 @default.
- W4287243477 hasConceptScore W4287243477C119857082 @default.
- W4287243477 hasConceptScore W4287243477C13280743 @default.
- W4287243477 hasConceptScore W4287243477C134306372 @default.
- W4287243477 hasConceptScore W4287243477C138885662 @default.
- W4287243477 hasConceptScore W4287243477C153180895 @default.
- W4287243477 hasConceptScore W4287243477C154945302 @default.
- W4287243477 hasConceptScore W4287243477C165696696 @default.
- W4287243477 hasConceptScore W4287243477C177148314 @default.
- W4287243477 hasConceptScore W4287243477C185798385 @default.
- W4287243477 hasConceptScore W4287243477C205649164 @default.
- W4287243477 hasConceptScore W4287243477C2776401178 @default.
- W4287243477 hasConceptScore W4287243477C2777212361 @default.
- W4287243477 hasConceptScore W4287243477C33923547 @default.
- W4287243477 hasConceptScore W4287243477C38652104 @default.
- W4287243477 hasConceptScore W4287243477C41008148 @default.
- W4287243477 hasConceptScore W4287243477C41895202 @default.
- W4287243477 hasConceptScore W4287243477C63584917 @default.
- W4287243477 hasConceptScore W4287243477C89600930 @default.
- W4287243477 hasLocation W42872434771 @default.
- W4287243477 hasOpenAccess W4287243477 @default.
- W4287243477 hasPrimaryLocation W42872434771 @default.
- W4287243477 hasRelatedWork W12783365 @default.
- W4287243477 hasRelatedWork W12820539 @default.
- W4287243477 hasRelatedWork W13815759 @default.
- W4287243477 hasRelatedWork W14789944 @default.
- W4287243477 hasRelatedWork W14828854 @default.
- W4287243477 hasRelatedWork W2873872 @default.
- W4287243477 hasRelatedWork W4061672 @default.
- W4287243477 hasRelatedWork W728297 @default.
- W4287243477 hasRelatedWork W7724241 @default.
- W4287243477 hasRelatedWork W8597964 @default.
- W4287243477 isParatext "false" @default.
- W4287243477 isRetracted "false" @default.
- W4287243477 workType "article" @default.