Table 3.

Results of 2D chemical structure image recognition on the test set

Thresholds Number of	Structures (% to the total)	Average Similarity Score
Cannot find the InChI	9 (4.41%)	-
T > 70%	72 (35.29%)	91.42
T > 80%	61 (29.90%)	94.43
T > 90%	44 (21.57%)	98.30
Identical structure	28 (13.73%)	100.00
Total mapped structure	144 (70.59%)	71.86

CACTVS script computed structure similarity between ground truth and regenerated structures based on standard InChI. In total 204 structures from PubChem were downloaded as the ground truth.