I have three questions about the training details

[clovaai]

I have three questions about the training details

[8uos]

Hi,
Q1: We just eliminated the duplicated fonts.
Q2: Right.
Q3: If you want to use the model weight that we uploaded, you should use our decompose json file.
New characters can be added to the file, but please remind that the defined component ids should be preserved.
If you are training a new model, you can build your own decompose json file.
Its format is a dictionary: {character name : [list of corresponding component ids]}.
The "character name" should be identical to the character name used in your lmdb dataset and meta files.
In this case, using different component ids ours is okay(In our case, 0 indicates ⺁, 1 indicates ⺌ ...).

[clscy]

ok, thank you.