This paper provides the first real-world evaluation of using visual (RGB camera) data and machine learning for proactively predicting millimeter wave (mmWave) dynamic link blockages before they happen. Proactively predicting line-of-sight (LOS) link blockages enables mmWave/sub-THz networks to make proactive network management decisions, such as proactive beam switching and hand-off) before a link failure happens. This can significantly enhance the network reliability and latency while efficiently utilizing the wireless resources. To evaluate this gain in reality, this paper (i) develops a computer vision based solution that processes the visual data captured by a camera installed at the infrastructure node and (ii) studies the feasibility of the proposed solution based on the large-scale real-world dataset, DeepSense 6G, that comprises multi-modal sensing and communication data. Based on the adopted real-world dataset, the developed solution achieves ≈90% accuracy in predicting blockages happening within the future 0.1s and ≈80% for blockages happening within 1s, which highlights a promising solution for mmWave/sub-THz communication networks.