Several Issues about unsupervised segmentation part.

[RealNewNoob]

I am trying to reproduce your segmentation research recently. However, I have some problems.
Firstly, I noticed that in the paper, you mentioned lamb was set as 1.5 for COCO3 and PostDam3 and 1.0 for other experiments. However, when I checked the setting for COCO3, I notice the lamd_A = 1.0 and lamd_B = 1.5, and I am a little confused.
In your shared model, I also noticed that take 555 for example, the num_epoches = 4800, while the loss data seems to have 109 rows in total. Does this mean you record it every 4800/109 epochs?
Another question is that shouldn't the minimum of loss _no_lamd be -1 since it is -1*mutual_information? Why the loss drops to a little higher than -1.5? Is the ideal minimum loss affected by the number of clusters?