Matches in SemOpenAlex for { <https://semopenalex.org/work/W2981352086> ?p ?o ?g. }
- W2981352086 abstract "We propose a novel image sampling method for differentiable image transformation in deep neural networks. The sampling schemes currently used in deep learning, such as Spatial Transformer Networks, rely on bilinear interpolation, which performs poorly under severe scale changes, and more importantly, results in poor gradient propagation. This is due to their strict reliance on direct neighbors. Instead, we propose to generate random auxiliary samples in the vicinity of each pixel in the sampled image, and create a linear approximation with their intensity values. We then use this approximation as a differentiable formula for the transformed image. We demonstrate that our approach produces more representative gradients with a wider basin of convergence for image alignment, which leads to considerable performance improvements when training networks for registration and classification tasks. This is not only true under large downsampling, but also when there are no scale changes. We compare our approach with multi-scale sampling and show that we outperform it. We then demonstrate that our improvements to the sampler are compatible with other tangential improvements to Spatial Transformer Networks and that it further improves their performance." @default.
- W2981352086 created "2019-11-01" @default.
- W2981352086 creator A5002942261 @default.
- W2981352086 creator A5037094498 @default.
- W2981352086 creator A5049610895 @default.
- W2981352086 creator A5049893251 @default.
- W2981352086 creator A5050009113 @default.
- W2981352086 date "2019-10-01" @default.
- W2981352086 modified "2023-10-14" @default.
- W2981352086 title "Linearized Multi-Sampling for Differentiable Image Transformation" @default.
- W2981352086 cites W1903029394 @default.
- W2981352086 cites W2035379092 @default.
- W2981352086 cites W2097073572 @default.
- W2981352086 cites W2097915756 @default.
- W2981352086 cites W2112796928 @default.
- W2981352086 cites W2117876524 @default.
- W2981352086 cites W2125188192 @default.
- W2981352086 cites W2263386426 @default.
- W2981352086 cites W2340296683 @default.
- W2981352086 cites W2412782625 @default.
- W2981352086 cites W2423984454 @default.
- W2981352086 cites W2520707372 @default.
- W2981352086 cites W2562066862 @default.
- W2981352086 cites W2601564443 @default.
- W2981352086 cites W2737590896 @default.
- W2981352086 cites W2777357193 @default.
- W2981352086 cites W2781155895 @default.
- W2981352086 cites W2798450692 @default.
- W2981352086 cites W2963150697 @default.
- W2981352086 cites W2963202462 @default.
- W2981352086 cites W2963635628 @default.
- W2981352086 cites W2963758027 @default.
- W2981352086 cites W2963857746 @default.
- W2981352086 cites W2981799316 @default.
- W2981352086 cites W3003662786 @default.
- W2981352086 doi "https://doi.org/10.1109/iccv.2019.00308" @default.
- W2981352086 hasPublicationYear "2019" @default.
- W2981352086 type Work @default.
- W2981352086 sameAs 2981352086 @default.
- W2981352086 citedByCount "19" @default.
- W2981352086 countsByYear W29813520862019 @default.
- W2981352086 countsByYear W29813520862020 @default.
- W2981352086 countsByYear W29813520862021 @default.
- W2981352086 countsByYear W29813520862022 @default.
- W2981352086 countsByYear W29813520862023 @default.
- W2981352086 crossrefType "proceedings-article" @default.
- W2981352086 hasAuthorship W2981352086A5002942261 @default.
- W2981352086 hasAuthorship W2981352086A5037094498 @default.
- W2981352086 hasAuthorship W2981352086A5049610895 @default.
- W2981352086 hasAuthorship W2981352086A5049893251 @default.
- W2981352086 hasAuthorship W2981352086A5050009113 @default.
- W2981352086 hasBestOaLocation W29813520862 @default.
- W2981352086 hasConcept C106131492 @default.
- W2981352086 hasConcept C110384440 @default.
- W2981352086 hasConcept C11413529 @default.
- W2981352086 hasConcept C115961682 @default.
- W2981352086 hasConcept C134306372 @default.
- W2981352086 hasConcept C137800194 @default.
- W2981352086 hasConcept C140779682 @default.
- W2981352086 hasConcept C153180895 @default.
- W2981352086 hasConcept C154945302 @default.
- W2981352086 hasConcept C160633673 @default.
- W2981352086 hasConcept C202615002 @default.
- W2981352086 hasConcept C205203396 @default.
- W2981352086 hasConcept C27405340 @default.
- W2981352086 hasConcept C31972630 @default.
- W2981352086 hasConcept C33923547 @default.
- W2981352086 hasConcept C41008148 @default.
- W2981352086 hasConcept C50644808 @default.
- W2981352086 hasConcept C9417928 @default.
- W2981352086 hasConceptScore W2981352086C106131492 @default.
- W2981352086 hasConceptScore W2981352086C110384440 @default.
- W2981352086 hasConceptScore W2981352086C11413529 @default.
- W2981352086 hasConceptScore W2981352086C115961682 @default.
- W2981352086 hasConceptScore W2981352086C134306372 @default.
- W2981352086 hasConceptScore W2981352086C137800194 @default.
- W2981352086 hasConceptScore W2981352086C140779682 @default.
- W2981352086 hasConceptScore W2981352086C153180895 @default.
- W2981352086 hasConceptScore W2981352086C154945302 @default.
- W2981352086 hasConceptScore W2981352086C160633673 @default.
- W2981352086 hasConceptScore W2981352086C202615002 @default.
- W2981352086 hasConceptScore W2981352086C205203396 @default.
- W2981352086 hasConceptScore W2981352086C27405340 @default.
- W2981352086 hasConceptScore W2981352086C31972630 @default.
- W2981352086 hasConceptScore W2981352086C33923547 @default.
- W2981352086 hasConceptScore W2981352086C41008148 @default.
- W2981352086 hasConceptScore W2981352086C50644808 @default.
- W2981352086 hasConceptScore W2981352086C9417928 @default.
- W2981352086 hasLocation W29813520861 @default.
- W2981352086 hasLocation W29813520862 @default.
- W2981352086 hasOpenAccess W2981352086 @default.
- W2981352086 hasPrimaryLocation W29813520861 @default.
- W2981352086 hasRelatedWork W1519931592 @default.
- W2981352086 hasRelatedWork W2037074416 @default.
- W2981352086 hasRelatedWork W2359696437 @default.
- W2981352086 hasRelatedWork W2361821952 @default.
- W2981352086 hasRelatedWork W2367652182 @default.
- W2981352086 hasRelatedWork W2372506239 @default.
- W2981352086 hasRelatedWork W2390710228 @default.
- W2981352086 hasRelatedWork W3080827050 @default.