Dear Authors,

Thanks for your attention. I have met some problem on replicating the results of forward facing dataset. I have trained a VM-96 model with the configuration under configs/flower.txt. The PSNR I get is around $28.2$ ~ $28.3$, which has a gap against $28.6$ reported by the paper. The depth maps rendered also contain more floater and aren't as clean as the results you provided. Would you mind suggesting whether the configs/flower.txt is the correct experiment setting to replicate your results? If not, how could I correctly configure the experiment to replicate the results?

Anyway, I'm really impressed by your marvelous work, as well as the clean implementation :)