In addition to coarse-grained received signal strength indicator (RSSI) measurements and fine-grained channel state information (CSI), a mid-grained channel measurement — spatial beam signal-to-noise ratios (SNRs) — that are inherently available during the millimeter wave (mmWave) beam training as defined in mmWave fifth-generation (5G) and IEEE 802.11ad/ay standards, were recently utilized for fingerprintingbased indoor localization. In this paper, we extend the beam SNR fingerprinting-based indoor localization to more challenging scenarios in non-line-of-sight (NLOS) propagation. Particularly, multi-channel beam covariance matrix (BCM) images are used as the fingerprinting signature and fed into a beam covariance learning (BCL) network to identify the position and estimate the coordinate. Using our in-house testbed with commercial off-theshelf (COTS) 60-GHz WiFi routers, real-world mmWave BCMs are fingerprinted in several NLOS locations-of-interest in an enclosed L-shape conference room. Given a fingerprinting gridsize of 30 cm, preliminary performance evaluation shows the position classification accuracy can be above 90% using classical classification methods and a coordinate estimation error around 11 cm with the BCL approach.