This dataset provides the code, trained models, and supporting data needed to reproduce the work presented in Jose et al. 2026 on transferable deep-learning-based OH-LIF segmentation. The study trains a segmentation model on synthetically generated, auto-labelled OH-LIF images derived from a single source flame and validates its drop-in transferability on six unseen target flames covering multiple facilities and imaging conditions. In addition, the work introduces an in-situ physical validation metric for quantifying segmentation accuracy without manual ground-truth labels and uses the final model to extract flame-front statistics from OH-LIF data.