Table 2.
Benchmark results for datasets with added distortions, such as mild shearing and rotation—performance of each model/tool on each dataset
| Benchmark results for datasets with distortions | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JPO (dist) | CLEF (dist) | USPTO (dist) | UOB (dist) | USPTO_big (dist) | Indigo (dist) | DECIMER-Test augmented | ||||||||
| Pi | T | Pi | T | Pi | T | Pi | T | Pi | T | Pi | T | Pi | T | |
| OSRA | 38% | 0.70 | 19% | 0.66 | 7% | 0.60 | 61% | 0.90 | 0.01% | 0.13 | 0.42% | 0.16 | 2% | 0.15 |
| MolVec | 41% | 0.80 | 21% | 0.66 | 26% | 0.71 | 63% | 0.92 | 0.02% | 0.14 | 0.48% | 0.07 | 1% | 0.12 |
| Imago | 23% | 0.47 | 33% | 0.65 | 51% | 0.81 | 34% | 0.64 | 0% | 0.08 | 0.01% | 0.20 | 0.15% | 0.10 |
| Img2Mol | 15% | 0.67 | 15% | 0.80 | 21% | 0.83 | 70% | 0.94 | 1% | 0.56 | 15% | 0.54 | 1% | 0.60 |
| SwinOCSR | 7% | 0.71 | 21% | 0.81 | 23% | 0.87 | 6% | 0.95 | 0% | 0.38 | 0.01% | 0.38 | 0.18% | 0.36 |
| MolScribe | 52% | 0.93 | 73% | 0.89 | 75% | 0.99 | 86% | 0.99 | 78% | 0.95 | 34% | 0.64 | 9% | 0.53 |
| DECIMER | 62% | 0.93 | 72% | 0.96 | 61% | 0.96 | 86% | 0.98 | 57% | 0.96 | 51% | 0.97 | 90% | 0.99 |
The performance is described as the proportion of occurrences of identical predictions Pi and the average Tanimoto similarity T.
The best result for each metric on each dataset is marked in bold.