search

clovaai / fewshot-font-generation

The unified repository for few-shot font generation methods. This repository includes FUNIT (ICCV'19), DM-Font (ECCV'20), LF-Font (AAAI'21) and MX-Font (ICCV'21).
Other
211 stars 37 forks source link

Some problems with training and validation #16

Open zi1zi opened 1 year ago

zi1zi commented 1 year ago

Thanks for the great work, I have the following questions

  1. I have trained 30 fonts, some images of the training process are as follows, I don’t think 65w iter is needed, as shown in the figure 511e233eddf26c1bc4f702f6e257f7bc
  2. Verification in training feels better than direct verification，The picture above is the verification during training, and the picture below is the direct verification 0200000-unseen_chars 6570e0afb2350a46a85b4b16c81cd94e 3.The overall effect is not very good, I have the following guesses: (1) There are few types of training fonts or the number of characters in each font is small (I use 800) (2) The number of training 20w iter is too small (3) There is an error in the use of the verification code. I use inference.py for verification.But I don't know what's the difference between the two image