Spaces:
Runtime error
Runtime error
File size: 5,001 Bytes
06142a4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 |
import json
import sys
import os
import argparse
import torch
import torch.nn as nn
import torchvision
from tqdm import tqdm
sys.path.append(os.getcwd())
from src.Text_Recognization.text_recognization import *
from src.Text_Recognization.prepare_dataset import *
from src.Text_Recognization.dataloader import *
device = 'cuda' if torch.cuda.is_available() else 'cpu'
def load_json_config(config_path):
with open(config_path, "r") as f:
config = json.load(f)
return config
def evaluate(model, dataloader, criterion, device):
model.eval()
losses = []
with torch.no_grad():
for images, labels, labels_len in dataloader:
images = images.to(device)
labels = labels.to(device)
outputs = model(images)
logits_lens = torch.full(
size=(outputs.size(1), ),
fill_value=outputs.size(0),
dtype=torch.long
).to(device)
loss = criterion(outputs, labels, logits_lens, labels_len)
losses.append(loss.item())
eval_loss = sum(losses) / len(losses)
return eval_loss
def training_loop(model, train_loader, val_loader, learning_rate, epochs, optimizer, criterion, scheduler, device):
model.to(device)
train_losses = []
val_losses = []
for epoch in range(epochs):
model.train()
batch_losses = []
for images, labels, labels_len in tqdm(train_loader):
images = images.to(device)
labels = labels.to(device)
optimizer.zero_grad()
outputs = model(images)
logits_lens = torch.full(
size=(outputs.size(1), ),
fill_value=outputs.size(0),
dtype=torch.long
).to(device)
loss = criterion(outputs, labels, logits_lens, labels_len)
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), 5)
optimizer.step()
batch_losses.append(loss.item())
train_loss = sum(batch_losses) / len(batch_losses)
train_losses.append(train_loss)
val_loss = evaluate(model, val_loader, criterion, device)
val_losses.append(val_loss)
print(f"epoch: {epoch+1}/{epochs}\ttrain_loss:{train_loss}\tval_loss:{val_loss}")
scheduler.step()
return train_losses, val_losses
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--root_path', type=str, default=os.getcwd(), help='Path to the root directory')
parser.add_argument('--checkpoints_path', type=str, default=os.path.join(os.getcwd(), 'checkpoints'), help='Path to the checkpoint directory')
args = parser.parse_args()
config_path = 'src/config.json'
dataset_path = os.path.join(args.root_path, 'Dataset')
config = load_json_config(config_path)
# dictionary char and idx
char_to_idx, idx_to_char = build_vocab(dataset_path)
# model
model = CRNN(vocab_size=config['vocab_size'], hidden_size=config['CRNN']['hidden_size'], n_layers=config['CRNN']['n_layers'])
# dataloader
train_loader, val_loader, test_loader = get_dataloader()
# define hyper parammeters
criterion = nn.CTCLoss(
blank=char_to_idx[config['blank_char']],
zero_infinity=True,
reduction='mean'
)
optimizer = torch.optim.Adam(
model.parameters(),
lr=config['CRNN']['learning_rate'],
weight_decay=config['CRNN']['weight_decay']
)
scheduler = torch.optim.lr_scheduler.StepLR(
optimizer=optimizer,
step_size=config['CRNN']['scheduler_step_size'],
gamma=0.1
)
# training loop
train_losses, val_losses = training_loop(
model=model,
train_loader=train_loader,
val_loader=val_loader,
learning_rate=config['CRNN']['learning_rate'],
epochs=config['CRNN']['epochs'],
optimizer=optimizer,
criterion=criterion,
scheduler=scheduler,
device=device
)
# save model
if not os.path.exists(args.checkpoints_path):
os.makedirs(args.checkpoints_path)
os.makedirs(os.path.join(args.checkpoints_path, 'losses'))
torch.save(model.state_dict(), os.path.join(args.checkpoints_path, 'crnn.pt'))
# draw losses
fig, axis = plt.subplots(1, 2, figsize=(8, 8))
axis[0].plot(train_losses, label='train_loss')
axis[0].set_xlabel('Epochs')
axis[0].set_ylabel('Loss')
axis[0].axis('off')
axis[0].legend()
axis[1].plot(val_losses, label='val_loss')
axis[1].set_xlabel('Epochs')
axis[1].set_ylabel('Loss')
axis[1].axis('off')
axis[1].legend()
plt.savefig(os.path.join(args.checkpoints_path, 'losses', 'losses.png'))
if __name__ == '__main__':
main() |