Skip to content

Latest commit

 

History

History
87 lines (68 loc) · 3.52 KB

LF-Font.md

File metadata and controls

87 lines (68 loc) · 3.52 KB
Raw

Train

Modify the configuration file ("cfgs/LF/p1/train.yaml", "cfgs/LF/p2/train.yaml")

You can set these values by giving command-line arguments like argparse, not modifying this configuration file directly. For the detailed description, please refer here.

  • use_ddp: Whether to use DataDistributedParallel. Set True to this to use multi-gpu.

  • port: The port for the DataDistributedParallel training.

  • decomposition: The location of decomposition rule file.

  • primals: The location of primals list file.

  • trainer: (leave blank)

    • resume: Path to the checkpoint to resume from.
    • work_dir: Path to save the checkpoints, the validation images, and log.

  • dset: (leave blank)

    • train: (leave blank)
      • source_path : path to the source font or source directory to use for the validation.
      • source_ext: extension of the source data.
        • If you are using a ttf file, set this to "ttf".
        • If you are using image files, set this to their extension ("png", "jpg" ...).

  • gen (leave blank) (only in "cfgs/LF/p2/train.yaml")

    • emb_dim: the dimension of style and component factors. (only in "cfgs/LF/p2/train.yaml")

Phase 1 training

python train_LF.py cfgs/LF/p1/train.yaml cfgs/data/train/custom.yaml --phase 1 --work_dir(optional) path/to/save/outputs

-g, -n, -nr, -p are arguments for the DistributedDataParallel training. You do not need to give these arguments if you are using a single GPU.

  • arguments
    • path/to/config (first argument): path to configration file.
      • Multiple values are allowed but the first one should locate in cfgs/LF.
    • -g : number of gpus to use for the training.
    • -n : number of nodes to use for the training.
    • -nr : the ranking of current node within the nodes.
    • -p : the port to use for the DistributedDataParallel training.
    • --work_dir : path to save outputs. The trainer.work_dir in the configuration file will be overwrited to this value.

Phase 2 training

python train_LF.py cfgs/LF/p2/train.yaml cfgs/data/train/custom.yaml --resume path/to/phase1/weights --phase 2 --work_dir(optional) path/to/save/outputs
  • arguments
    • path/to/config (first argument, multiple values are allowed): path to configration file.
      • Multiple values are allowed but the first one should locate in cfgs/LF/p2.
    • --resume : path to the weight which saved by phase 1 training.
    • --phase : The training phase. 1 or 2 is available.
    • --work_dir(optional) : path to save outputs. The trainer.work_dir in the configuration file will be overwrited to this value.

Evaluate

1. Modify the configuration file ("cfgs/LF/p2/eval.yaml")

All the arguments should be identical to the arguments used for the training the weight to evaluate.

  • decomposition: The location of decomposition rule file.
  • primals: The location of primals list file.

2. Run evaluation

python inference.py cfgs/LF/p2/eval.yaml cfgs/data/eval/chn_ttf.yaml \
--model LF \
--weight weights/LF_chn.pth \
--result_dir ./result/LF
  • arguments
    • path/to/config (first argument): path to configration files.
      • Multiple values are allowed but the first one should locate in cfgs/LF/p2.
    • --model : The model to evaluate. DM, LF, MX and FUNIT are available.
    • --weight: The weight to evaluate.
    • --result_dir: Path to save generated images.
    • --n_ref: The number of reference characters to use for the generation.