DART — datasets, model weights, and chip design

Supporting data for the paper “DART: A design-aware microfluidic chip paradigm for real-time live-cell image analysis” (Seiffarth et al., 2026). This record bundles the datasets, the trained fiducial-marker detection model weights, and the Swiss Army Knife (SAK) chip design file used to develop and validate the DART paradigm. The DART software is archived separately (see Software below). Contents. The record is a flat list of files. Each dataset is a .zip that unpacks to its own folder tree; the chip design and model weights are provided standalone. dart_train_test_dataset.zip (~0.88 GB) — Annotated phase-contrast images used to train / validate / test the YOLO fiducial-marker detector (472 cross + 470 circle annotations; see Table S2). calibration_data_full.zip (~9.0 GB) — Coarse-alignment validation dataset: images + stage coordinates used to quantify RoI-localization accuracy (median 10.46 µm, 90th percentile 18.39 µm; Figure S3). live_cell_experiment.zip (~3.2 GB) — Live-cell C. glutamicum imaging dataset (1,739 images across the SAK geometries) used for end-to-end pipeline validation (Figure 5, Table S5). Unpacks to one folder per chamber geometry (Small Chambers, Small Chambers + Pillar, Big Chambers, Big Chambers + Pillars, Open Chambers, Open Chambers + Structures, Mother Machines; multi-page .tif stacks) and also includes a copy of SAK_blueprint.cif. SAK_blueprint.cif (~0.4 MB) — The SAK chip CAD design file (CleWin CIF), provided standalone so adopters can download it directly without fetching the imaging data. v26_detect_s_imgsz1280.pt (~20 MB) — Trained YOLO marker-detection weights (YOLOv26-s, detection, 1280 px). Training / runtime details in the paper (Table S6). Mapping to the paper. Marker detection (Results: Real-time fiducial marker detection; Table S4) → dart_train_test_dataset.zip, v26_detect_s_imgsz1280.pt Coarse alignment / RoI localization (Results: Coarse alignment; Figure S3) → calibration_data_full.zip Fine alignment + live-cell validation (Results: Live-cell experiment validation; Figure 5, Table S5) → live_cell_experiment.zip Chip design / adoption (Methods: Converting a microfluidic design into a DART chip) → SAK_blueprint.cif Software (archived separately). The DART software for coarse and fine alignment is available at https://github.com/SMLCI/DART-MLCI (also on PyPI). Exact runtime versions are listed in Table S6 of the paper (Python 3.10.6, PyTorch 2.9.1 / CUDA 12.8, Cellpose-SAM 4.0.5, Ultralytics 8.3.249, Kornia 0.8.2, NumPy 2.2.6, OpenCV 4.12.0, scikit-image 0.25.2, Shapely 2.1.2, Rasterio 1.4.4).