|
|
import torch |
|
|
|
|
|
|
|
|
|
|
|
EXP_NAME = "IAM-339-15-E3D3-LR0.00005-bs8"; RESUME = False |
|
|
|
|
|
DATASET = 'IAM' |
|
|
if DATASET == 'IAM': |
|
|
DATASET_PATHS = 'files/IAM-32.pickle' |
|
|
NUM_WRITERS = 339 |
|
|
if DATASET == 'CVL': |
|
|
DATASET_PATHS = 'files/CVL-32.pickle' |
|
|
NUM_WRITERS = 283 |
|
|
ENGLISH_WORDS_PATH = 'files/english_words.txt' |
|
|
|
|
|
|
|
|
|
|
|
IMG_HEIGHT = 32 |
|
|
resolution = 16 |
|
|
batch_size = 8 |
|
|
NUM_EXAMPLES = 15 |
|
|
TN_HIDDEN_DIM = 512 |
|
|
TN_DROPOUT = 0.1 |
|
|
TN_NHEADS = 8 |
|
|
TN_DIM_FEEDFORWARD = 512 |
|
|
TN_ENC_LAYERS = 3 |
|
|
TN_DEC_LAYERS = 3 |
|
|
ALPHABET = 'Only thewigsofrcvdampbkuq.A-210xT5\'MDL,RYHJ"ISPWENj&BC93VGFKz();#:!7U64Q8?+*ZX/%' |
|
|
VOCAB_SIZE = len(ALPHABET) |
|
|
G_LR = 0.00005 |
|
|
D_LR = 0.00005 |
|
|
W_LR = 0.00005 |
|
|
OCR_LR = 0.00005 |
|
|
EPOCHS = 100000 |
|
|
NUM_CRITIC_GOCR_TRAIN = 2 |
|
|
NUM_CRITIC_DOCR_TRAIN = 1 |
|
|
NUM_CRITIC_GWL_TRAIN = 2 |
|
|
NUM_CRITIC_DWL_TRAIN = 1 |
|
|
NUM_FID_FREQ = 100 |
|
|
|
|
|
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
|
|
IS_SEQ = True |
|
|
NUM_WORDS = 3 |
|
|
if not IS_SEQ: NUM_WORDS = NUM_EXAMPLES |
|
|
IS_CYCLE = False |
|
|
IS_KLD = False |
|
|
ADD_NOISE = False |
|
|
ALL_CHARS = False |
|
|
SAVE_MODEL = 5 |
|
|
SAVE_MODEL_HISTORY = 100 |
|
|
|
|
|
def init_project(): |
|
|
import os, shutil |
|
|
if not os.path.isdir('saved_images'): os.mkdir('saved_images') |
|
|
if os.path.isdir(os.path.join('saved_images', EXP_NAME)): shutil.rmtree(os.path.join('saved_images', EXP_NAME)) |
|
|
os.mkdir(os.path.join('saved_images', EXP_NAME)) |
|
|
os.mkdir(os.path.join('saved_images', EXP_NAME, 'Real')) |
|
|
os.mkdir(os.path.join('saved_images', EXP_NAME, 'Fake')) |
|
|
|
|
|
|
