Fig. 8. Representative examples of cyclic (top) and acyclic (bottom) molecules from the hand-drawn hydrocarbon test set and their corresponding predictions from the ensemble. The input image is presented next to the predictions; the number of votes for each predicted molecule, V, is shown, along with if the molecule was recognised correctly. Predictions of invalid SMILES strings are shown as N/A. Hydrocarbons that are recognised correctly by one of the models predictions are outlined in green, and those that fail to be predicted correctly are outlined in red.
