CucurPollen

A microscopy image dataset of pollen germination and tube growth from melon, watermelon, cucumber, and pumpkin cultivated under naturally occurring heat stress conditions. Beyond supporting computer vision research, the dataset is linked to two agronomic assays involving a diverse panel of cultivars, providing a phenotyping resource for studying pollen viability responses to warm cultivation conditions across cucurbit germplasm. The dataset comprises 3089 microscopy images accompanied by experimental metadata. A curated subset of 122 images includes instance-level annotations in COCO format distinguishing germinated grains, non-germinated grains, and pollen tubes, while a complementary subset of 639 images provides expert visual scores describing pollen abundance, germination level, and tube development. Additionally, 129 images are flagged for visual noise. Technical validation was performed by training and evaluating several semantic segmentation models on the annotated subset, demonstrating that the annotations support reliable automated pollen segmentation. The dataset provides a benchmark for developing and validating automated pollen analysis methodologies while supporting future studies on pollen performance under naturally occurring high-temperature conditions.