Should the l_weight value change according to the dataset?

[ba77-ku]

l_weight = [0.7,0.7,1.1,1.1,0.3,0.3,1.3] for bdcn

Should this value be calculated for each different dataset? Can you explain the method you used?

[xavysp]

Hi, we have not explored this yet, apparently yes. But changes on l_weight may change slightly the results. Even changes in WD or LR.

[ba77-ku]

Accually ı changed some layers and ı improve my prediction result accuracy. but ı think, how can ı change l_weight values? thank for your attention.

[xavysp]

In my case, as you see [0.7,0.7,1.1,1.1,0.3,0.3,1.3] I set them with larger weights in 3rd, 4th and the fusion output. Since those outputs, according to the observation has better edge predictions.