inhouse data for EpiDataset

This dataset comprises the in-house training data acquired and constructed for fine-tuning the EpiVision vision module in the accompanying manuscript. All samples were captured using bright-field microscopy and prepared via the epidermal peeling method. The dataset includes: 170 samples of dicotyledonous tomato (Solanum lycopersicum) 210 samples of monocotyledonous wheat (Triticum aestivum) To adapt to network computation, the acquired raw samples were subjected to non-overlapping cropping, generating a total of 530,470 standardized patches at a resolution of 512 × 512 pixels. A representative subset of patches was annotated using ISAT annotation software, comprising: 407 tomato patches and 58 wheat patches The annotated subset contains the following instance-level annotations: Cell Type Tomato Wheat Total Pavement cells 5,419 1,780 7,199 Stomatal complexes 1,471 426 1,897 Stomatal pores 629 51 680 This dataset supports the training and evaluation of the EpiVision model for instance segmentation of stomatal complexes, stomatal pores, and pavement cells.