Harmonizing Output Imbalance for semantic segmentation on extremely-imbalanced input data.
Jianye YiXiaopin ZhongWeixiang LiuZongze WuYuanlong DengPublished in: CoRR (2022)
Keyphrases
- input data
- semantic segmentation
- street scenes
- class imbalance
- superpixels
- label transfer
- scene classification
- conditional random fields
- input image
- weakly supervised
- data sets
- object categories
- image data
- data points
- image understanding
- object class
- face recognition
- pascal voc
- higher order
- object classes
- image set
- feature vectors
- multiscale
- segmentation algorithm
- active learning
- probabilistic model
- training data
- similarity measure