Background & Aim: Seismocardiography (SCG) enables non-invasive assessment of cardiac function by capturing chest vibrations associated with physiological events such as aortic valve opening (AO) and aortic valve closure (AC). Reliable detection of these fiducial points is essential for clinical applicability. A previously proposed deep learning-based algorithm (SeismoTracker) has shown promising results for beat-to-beat AO and AC detection but has only been validated on internal datasets. External validation is necessary to assess robustness and generalizability across different conditions. Therefore, this study aims to validate the algorithm on an independent dataset. Methods: An open-source dataset consisting of SCG recordings from 6 pigs undergoing controlled hypovolemia was used. The dataset consisted of 81983 individual heartbeats with mean number of heartbeats pr pig was 13663 (3072). Reference AO and AC timings were derived from the continuously recorded aortic pressure signal on beat-to-beat basis, while AO and AC timings also were extracted on beat-to-beat basis using SeismoTracker. Performance was evaluated using sensitivity, positive predictive value (PPV), and timing bias across all pigs. Results: The median (IQR) PPV was 0.968 (0.029) for AO and 0.883 (0.151) for AC, indicating a low rate of false positive detections. Sensitivity achieved median (IQR) values of 0.473 (0.337) and 0.458 (0.326) for AO and AC, respectively. The bias was 1.48 ms for AO and -11.57 ms for AC. Conclusion: The algorithm demonstrated robust performance in terms of detection precision on external data, supporting its generalizability. However, the relatively low sensitivity highlights the need for further development, particularly through inclusion of more diverse training data to improve detection completeness while maintaining high PPV.