Training script fonts are different from fonts used in training dataset

[khawar-islam]

When i have generated a synthetic training dataset, I have used different fonts style contained from different english and korean like below:

Training dataset fonts

During training, the script is using different fonts like FreeMono.ttf,FreeSans.ttf,FreeSerif.ttf.

Namespace(amp=False, arch='master', batch_size=16, device=None, epochs=50, find_lr=False, font='FreeMono.ttf,FreeSans.ttf,FreeSerif.ttf', input_size=32, lr=0.001, max_chars=12, min_chars=1, name=None, pretrained=False, push_to_hub=False, resume=None, sched='cosine', show_samples=False, test_only=False, train_path='/media/cvpr/CM_1/CC/dataset_KR/dataset/train', train_samples=1000, val_path='/media/cvpr/CM_1/CC/dataset_KR/dataset/val', val_samples=20, vocab='korean', wb=False, weight_decay=0, workers=None)

Are there any problem while training master model? Is this right way?

[felixdittrich92]

Hi @khawar-islam you need to pass the list of fonts you want to use as arg to the training script (be aware that the fonts are supported by the vocab characters)

parser.add_argument(
        "--font", type=str, default="FreeMono.ttf,FreeSans.ttf,FreeSerif.ttf", help="Font family to be used"
    )