Table 4 Comparing the encoder–decoder- and transformer-based approach with a 1 million images test dataset
From: DECIMER 1.0: deep learning for chemical image recognition using transformers
Metrics | Encoder–decoder | Transformer | ||
|---|---|---|---|---|
InceptionV3 | EfficientNet-B3 | InceptionV3 | EfficientNet-B3 | |
Average training time per epoch | 7 min 34 s | 8 min 57 s | 8 min 33 s | 9 min 27 s |
Tanimoto | 0.5459 | 0.6345 | 0.8764 | 0.9371 |
Tanimoto 1.0 | 1.41% | 7.03% | 55.29% | 74.57% |