Fig. 7
From: Learning from label proportions for SAR image classification
Influence of label proportion errors on classification accuracy using GL+LpcSVM for SAR images of Tianjin and Rosenheim areas. Noise standard deviation σ ranges 0.05, 0.10, 0.15, and 0.20. Note that σ=0 indicates n k =0, meaning no noises added in p k . The special case that no label proportions are provided in the training data (degraded to naive GL) is also considered, denoted by σ= max