A Continuous Multimodal Wearable Dataset of Synchronized Cardiac, Kinematic and Heat-Production Signals from Growing Pigs in Open-Circuit Respirometry Chambers (Porcine-CaMo January-2025 Cohort, 3 Chambers / 9 Pigs / 22-Day Trial Window / 45 Pig-Days, Accelerometer-Free Configuration with Professionally-Audited ECG Annotation Training Subset)

Raw signals are organized by chamber and date; each recording session produces paired Raw_*.txt (ECG) and AngleTemp_*.txt (gyroscope + temperature) files. Heat production was computed by OpenCalori-Swine (github.com/zengzhengcheng/OpenCalori-Swine<br>Zenodo DOI: 10.5281/zenodo.20051163) via the Brouwer equation. ECG streams were processed with ECG TransUNet (github.com/zengzhengcheng/ECG-TransUNet<br>Zenodo DOI: 10.5281/zenodo.20051167) for automatic R-peak detection followed by adaptive statistical correction, producing one per-chamber R-peak label series.

Because of long retention before publication, the Perfect quality tier of this cohort's ECG was lost and is not part of this DOI; only the Corrected tier is released. As a high-accuracy equivalent subset, this dataset additionally publishes a dedicated top-level ECGModelTrainData/ directory containing ≈ 20 ECG training segments from B1 + B2 — this subset was reviewed by professionally hired human annotators and serves as the base-training corpus of ECG TransUNet (the correction fine-tuning corpus is excellent_data.hdf in the ECG-TransUNet repository root, Git LFS-tracked, independent of and not derived from this directory).

Per-chamber 1 Hz integrated signals (*heart.csv + *move_v2.csv), the 5-min ML-ready feature table (A1B1B2_features.csv, 6,812 rows × 158 columns, plus per-chamber slices), five runnable v4 reference training scripts and four work-summary documents are bundled under TrainDataAndCode/. HP prediction models are trained per cohort separately; this DOI ships training features and reference scripts but does not ship pre-trained model weights.

The HRV-processing pipeline, the gyroscope-only IMU pipeline and the ECG TransUNet ONNX inference weights are publicly open-sourced at github.com/zengzhengcheng/SwineSync-OpenSource<br>Zenodo DOI: 10.5281/zenodo.20051135; the heart-rate annotation GUI (SwineSync Studio annotator) is retained as an internal product tool distributed only as a Windows executable. The release supports precision livestock farming, accelerometer-free metabolism–behaviour modelling, welfare phenotyping, ECG R-peak detection model training, and benchmarking of wearable animal-physiology algorithms.