We show the recommendation accuracy calculated as = # correctly recommended harmful and mislabelled training samples / # harmful and mislabelled training samples
| Mislabelling Ratio | CUB200 | CARS196 | InShop |
+-----------------------------+---------+----------+
| 1% | 84.62% | 100.00% | 96.52% |
+-----------------------------+---------+----------+
| 5% | 88.00% | 91.30% | 95.24% |
+-----------------------------+---------+----------+
| 10% | 86.36% | 97.44% | 97.74% |
+-----------------------------+---------+----------+
We show the recommendation accuracy calculated as = # correctly recommended harmful and mislabelled training samples / # harmful and mislabelled training samples
| Mislabelling Ratio | CUB200 | CARS196 | InShop |
+-----------------------------+---------+----------+
| 1% | 100.00%| 96.42% | 97.12% |
+-----------------------------+---------+----------+
| 5% | 100.00%| 91.89% | 95.17% |
+-----------------------------+---------+----------+
| 10% | 96.00% | 91.89% | 92.19% |
+-----------------------------+---------+----------+