In this paper, we test two anomaly detection methods based on Generative Adversarial Networks (GAN) on office monitoring including humans. GAN-based methods, especially those equipped with encoders and decoders, have shown impressive results in detecting new anomalies from images. We have been working on human monitoring in office environments with autonomous mobile robots and are motivated to incorporate the impressive, recent progress of GAN-based methods. Lawson et al.’s work tackled a similar problem of anomalous detection in an indoor, patrol trajectory environment with their patrolbot with a GAN-based method, though crucial differences such as the absence of humans exist for our purpose. We test a variant of their method, which we call FA-GAN here, as well as the cutting-edge method of GANomaly on our own robotic dataset. Motivated to employ such a method for a turnable Video Camera Recorder (VCR) placed at a fixed point, we also test the two methods for another dataset. Our experimental evaluation and subsequent analyses revealed interesting tendencies of the two methods including the effect of a missing normal image for GANomaly and their dependencies on the anomaly threshold.