From: DECIMER 1.0: deep learning for chemical image recognition using transformers
Metrics | Subset 1 | Subset 2 | Subset 3 | Subset 4 |
---|---|---|---|---|
Train data size | 921,600 | 10,240,000 | 15,360,000 | 35,002,240 |
Test data size | 102,400 | 1,024,000 | 1,536,000 | 3,929,093 |
Predictions with Tanimoto 1.0 | 74,176 | 899,941 | 1,398,028 | 3,790,273 |
Isomorphic predictions | 98.63% | 99.45% | 99.59% | 99.75% |
Non-isomorphic predictions | 1.37% | 0.55% | 0.41% | 0.25% |