Code details and clarification

[MihaiDavid05]

Hello @ccdatas,

I noticed that the alpha and beta parameters are not in the config, but I saw that they are set to 4, at least for KNet, as I was looking through the code in here and here for class and mask. However I see that there are 2 more multipliers, 12 and 2.5, for class and mask, respectively, which I cannot figure out what they are. Can you give me an insight? Or are they found just empirically and not specified in the paper?

Also, for the MaskRCNN version, could you please point out where is the alpha parameter basically integrated? As for the beta parameter, I found it to be 4 in here, but again there is a multiplier=5 in the weight formula, that I have no clue about.

Lastly, I did not figure out why there is a 0.35 * 0.35 multiplier for the negative samples in case of unsupervised input here, as it does not appear in the formula from the paper.

Thank you very much for your help!

[MihaiDavid05]

@ccdatas do you have any updates on these?

Thank you very much!

[ccdatas]

Hello, @MihaiDavid05
Some parameters are found just empirically.
For the MaskRCNN version, the alpha parameter is 4.
There is no special reason why negative samples have 0.35 * 0.35 multiplier, even if it is 1.0, the experimental effect is about the same.