TAGNet: Learning Configurable Context Pathways for Semantic Segmentation
| Autor: | Di Lin, Dingguo Shen, Yuanfeng Ji, Siting Shen, Mingrui Xie, Wei Feng, Hui Huang |
|---|---|
| Rok vydání: | 2023 |
| Předmět: | |
| Zdroj: | IEEE Transactions on Pattern Analysis and Machine Intelligence. 45:2475-2491 |
| ISSN: | 1939-3539 0162-8828 |
| DOI: | 10.1109/tpami.2022.3165034 |
| Popis: | State-of-the-art semantic segmentation methods capture the relationship between pixels to facilitate context exchange. Advanced methods utilize fixed pathways, lacking the flexibility to harness the most relevant context for each pixel. In this paper, we present Configurable Context Pathways (CCP), a novel scheme for establishing pathways for augmenting context. In contrast to previous methods, the pathways are learned, leveraging configurable contextual regions to form information flows between pairs of pixels. The regions are adaptively configured, driven by the relationships between remote pixels, spanning over the entire image space. Subsequently, the information flows along the pathways are gradually updated by the information provided by sequences of configurable regions, forming more powerful context. We extensively evaluate our method on competitive benchmarks, demonstrating that all of its components effectively improve the segmentation accuracy and help to surpass the state-of-the-art results. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|