Semantic Understanding of Scenes through the ADE20K Dataset

Semantic Understanding of Scenes through the ADE20K Dataset

16 Oct 2018 | Bolei Zhou · Hang Zhao · Xavier Puig · Tete Xiao · Sanja Fidler · Adela Barriuso · Antonio Torralba
The paper introduces the ADE20K dataset, a densely annotated dataset for scene understanding, which includes 20,210 images with detailed annotations of objects, parts, and attributes. The dataset covers a wide range of scenes and object categories, with an average of 19.5 instances and 10.5 object classes per image. Based on ADE20K, the authors construct benchmarks for scene parsing and instance segmentation, providing baseline performances and re-implementing state-of-the-art models. They also evaluate the impact of synchronized batch normalization and find that a reasonably large batch size is crucial for semantic segmentation performance. The networks trained on ADE20K are capable of segmenting a variety of scenes and objects. The paper further discusses the Places Challenges organized in 2016 and 2017, and explores applications such as hierarchical semantic segmentation and automatic image content removal.The paper introduces the ADE20K dataset, a densely annotated dataset for scene understanding, which includes 20,210 images with detailed annotations of objects, parts, and attributes. The dataset covers a wide range of scenes and object categories, with an average of 19.5 instances and 10.5 object classes per image. Based on ADE20K, the authors construct benchmarks for scene parsing and instance segmentation, providing baseline performances and re-implementing state-of-the-art models. They also evaluate the impact of synchronized batch normalization and find that a reasonably large batch size is crucial for semantic segmentation performance. The networks trained on ADE20K are capable of segmenting a variety of scenes and objects. The paper further discusses the Places Challenges organized in 2016 and 2017, and explores applications such as hierarchical semantic segmentation and automatic image content removal.
Reach us at info@study.space
[slides] Semantic Understanding of Scenes Through the ADE20K Dataset | StudySpace