Paper: Human-annotated label noise and their impact on ConvNets for remote sensing image scene classification (under review)
-
We obtained human annotations for the UC Merced dataset from 32 independent labelers.
-
Please find these human annotations with real-world label noise in the file 'human_annotations.xlsx'
- The first column stores the image names. These image names are the same as those in the UCMerced dataset.
- The remaining columns indicate human annotations from 32 independent labelers. Each column corresponds to human annotations from a single labeler, with the labeler index indicated in the first row as 'S*' (e.g., S01 and S02).
-
Here is the preprint version of our work.