Lawin Transformer: Improving Semantic Segmentation Transformer with Multi-Scale Representations via Large Window Attention.
Haotian YanChuang ZhangMing WuPublished in: CoRR (2022)
Keyphrases
- semantic segmentation
- multiscale
- superpixels
- street scenes
- conditional random fields
- natural images
- weakly supervised
- scene classification
- label transfer
- pixel level
- feature extraction
- scale space
- image representation
- level set
- object categories
- object classes
- object detection
- small number
- image features
- viewpoint
- pairwise