Partaker: Segmentation benchmark and U-Net training datasets

Datasets supporting the Partaker bacterial image analysis tool. Benchmark dataset: 45 phase-contrast frames with manually corrected ground-truth instance masks across four microfluidic chamber positions (P0, P4, P5, P6), sampled across early, intermediate, and late growth phases, with per-frame instance-level segmentation metrics. U-Net training dataset: 80 phase-contrast frames across four positions with binary and instance masks, boundary weight maps, and trained model weights (Keras .h5 and PyTorch .pt). Acquisition: Nikon Eclipse Ti2-E PFS, 100x oil-immersion APO objective, Dhyana 400D sCMOS camera, pixel size 0.0645 um/pixel. Two-strain E. coli co-culture in monolayer microfluidic chambers.