def main():
config = parse_arg()
if torch.cuda.is_available():
torch.backends.cudnn.benchmark = True
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = get_crnn(config).to(device)
if config.TRAIN.RESUME.IS_RESUME:
model_state_file = config.TRAIN.RESUME.FILE
if model_state_file != '' and os.path.exists(model_state_file):
print('loading pretrained model from %s' % model_state_file)
model.load_state_dict(torch.load(model_state_file))
criterion = torch.nn.CTCLoss(reduction='sum').to(device)
val_dataset = OcrDataset(config, is_train=False)
val_dataloader = data.DataLoader(
dataset=val_dataset,
batch_size=config.TEST.BATCH_SIZE_PER_GPU,
shuffle=config.TEST.SHUFFLE,
num_workers=config.WORKERS,
pin_memory=config.PIN_MEMORY,
)
converter = utils.strLabelConverter(alphabet)
acc = validate(config, val_dataloader, converter, model, criterion, device)
def __init__(self, model_path='./checkpoints/CRNN.pth'):
self.alphabet = ''.join([chr(uni) for uni in crnn_params.alphabet])
self.nclass = len(self.alphabet) + 1
self.model = CRNN(crnn_params.imgH, 1, self.nclass, 256)
self.use_gpu = torch.cuda.is_available()
if self.use_gpu:
self.model.cuda()
self.model.load_state_dict(torch.load(model_path))
for p in self.model.parameters():
p.requires_grad = False
self.model.eval()
self.converter = strLabelConverter(self.alphabet)
def __init__(self, model_path):
#def crnnSource(model_path, use_gpu=True):
alphabet = keys.alphabet # Chinese words
self.converter = crnn_utils.strLabelConverter(alphabet)
# note that in https://github.com/bear63/sceneReco support multi GPU.
# model = crnn.CRNN(32, 1, len(alphabet)+1, 256, 1).cuda()
self.model = crnn.CRNN(32, 1, len(alphabet) + 1, 256)
self.cpu_model = crnn.CRNN(32, 1, len(alphabet) + 1, 256)
if torch.cuda.is_available():
self.model = self.model.cuda()
print('loading pretrained model from %s' % model_path)
#model_path = './crnn/samples/netCRNN63.pth'
model_state_dict = torch.load(model_path)
self.model.load_state_dict(model_state_dict)
self.cpu_model.load_state_dict(model_state_dict)
def val(model, loader, criterion, device):
print('Start val')
for p in model.parameters():
p.requires_grad = False
model.eval()
loss_avg = utils.averager()
alphabet = ''.join([chr(uni) for uni in crnn_params.alphabet])
converter = utils.strLabelConverter(alphabet)
n_total = 0
n_correct = 0
preds = 0
# for i_batch, (image, label, index) in enumerate(loader):
for i_batch, (image, label, index) in tqdm(enumerate(loader),
total=len(loader),
desc='test model'):
image = image.to(device)
preds = model(image)
batch_size = image.size(0)
index = np.array(index.data.numpy())
text, length = converter.encode(label)
preds_size = torch.IntTensor([preds.size(0)] * batch_size)
cost = criterion(preds, text, preds_size, length) / batch_size
loss_avg.add(cost)
_, preds = preds.max(2)
preds = preds.transpose(1, 0).contiguous().view(-1)
sim_preds = converter.decode(preds.data, preds_size.data, raw=False)
for pred, target in zip(sim_preds, label):
if pred == target:
n_correct += 1
n_total += batch_size
raw_preds = converter.decode(preds.data, preds_size.data,
raw=True)[:crnn_params.n_test_disp]
for raw_pred, pred, gt in zip(raw_preds, sim_preds, label):
print('%-20s => %-20s, gt: %-20s' % (raw_pred, pred, gt))
accuracy = n_correct / float(n_total)
print('Test loss: %.6f, accuray: %.6f' % (loss_avg.val(), accuracy))
return accuracy
def __init__(self, *args, **kwargs):
super(ModelIinit, self).__init__(*args, **kwargs)
if self.model_params["model_type"] == "crnn_big_size":
self.model = crnn_big_size.CRNN(
nc=self.model_params["num_input_channels"],
nclass=self.nclass,
nh=self.model_params["hid_layer_size"])
self.converter = crnn_utils.strLabelConverter(
self.model_params["alphabet"])
self.criterion = CTCLoss(zero_infinity=True).to(
self.general_params["device"])
self.model.apply(self.weights_init)
'''load pretrained weigths'''
path_to_pretrained_model = self.model_params[
self.model_params["model_type"]]["path_pretrained"]
if path_to_pretrained_model and os.path.isfile(
path_to_pretrained_model):
print('loading pretrained model')
self.model.load_state_dict(path_to_pretrained_model)
self.model.to(self.general_params["device"])
self.model = torch.nn.DataParallel(self.model,
device_ids=range(
self.general_params["num_gpu"]))
'''optimizer initialise'''
if self.model_params["optimizer"] == "Adam":
self.optimizer = optim.Adam(self.model.parameters(),
lr=self.model_params["adam"]["lr"],
betas=(self.model_params["adam"]["lr"],
0.999))
elif self.model_params["optimizer"] == "adadelta":
self.optimizer = optim.Adadelta(self.model.parameters(),
lr=self.model_params["adam"]["lr"])
else:
self.optimizer = optim.RMSprop(self.model.parameters(),
lr=self.model_params["adam"]["lr"])
def train(model, loader, criterion, optimizer, iteration, device):
for p in model.parameters():
p.requires_grad = True
model.train()
loss_avg = utils.averager()
alphabet = ''.join([chr(uni) for uni in crnn_params.alphabet])
converter = utils.strLabelConverter(alphabet)
for i_batch, (image, label, index) in enumerate(loader):
image = image.to(device)
preds = model(image)
batch_size = image.size(0)
text, length = converter.encode(label)
preds_size = torch.IntTensor([preds.size(0)] * batch_size)
cost = criterion(preds.log_softmax(2), text, preds_size, length) / batch_size
model.zero_grad()
cost.backward()
optimizer.step()
loss_avg.add(cost)
if (i_batch+1) % crnn_params.displayInterval == 0:
theTime = datetime.datetime.now()
print('%s [%d/%d][%d/%d] Loss: %f' % (theTime, iteration, crnn_params.niter, i_batch, len(loader), loss_avg.val()))
loss_avg.reset()
model_path = './checkpoints/CRNN.pth'
alphabet = keys.alphabet
imgH = 32
imgW = 280
model = crnn.CRNN(imgH, 1, len(alphabet) + 1, 256)
if gpu:
model = model.cuda()
print('loading pretrained model from %s' % model_path)
if gpu:
model.load_state_dict( torch.load( model_path ) )
else:
model.load_state_dict(torch.load(model_path,map_location=lambda storage,loc:storage))
model.eval()
print('done')
print('starting...')
converter = crnn_utils.strLabelConverter(alphabet)
transformer = mydataset.resizeNormalize3((imgW, imgH))
def recognize_cv2_image(img):
img = cv2.cvtColor( img, cv2.COLOR_BGR2RGB )
image = Image.fromarray(np.uint8(img)).convert('L')
image = transformer( image )
if gpu:
image = image.cuda()
preds = model( image )
preds = F.log_softmax(preds,2)
conf, preds = preds.max( 2 )
preds = preds.transpose( 1, 0 ).contiguous().view( -1 )
preds_size = Variable( torch.IntTensor( [preds.size( 0 )] ) )
print('alphabet length : ', config.MODEL.NUM_CLASSES)
if torch.cuda.is_available():
torch.backends.cudnn.benchmark = True
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = get_crnn(config).to(device)
if args.model_path != '' and os.path.exists(args.model_path):
print('loading pretrained model from %s' % args.model_path)
model.load_state_dict(torch.load(args.model_path))
image = Image.open(args.image_name).convert("L")
w, h = image.size
new_w = int(w / h * config.MODEL.IMAGE_SIZE.H)
image = image.resize((new_w, config.MODEL.IMAGE_SIZE.H))
image = np.array(image).astype(np.float32)
image = (image / 255.0 - config.DATASET.MEAN) / config.DATASET.STD
image = np.expand_dims(image, axis=0)
image = np.expand_dims(image, axis=0)
image = torch.from_numpy(image)
converter = utils.strLabelConverter(alphabet)
model.eval()
with torch.no_grad():
image = image.to(device)
preds = model(image)
preds_size = torch.IntTensor([preds.size(0)] * image.size(0))
_, preds = preds.max(2)
preds = preds.transpose(1, 0).contiguous().view(-1)
sim_preds = converter.decode(preds.data, preds_size.data, raw=False)
print(sim_preds)
def main():
config = parse_arg()
output_dict = utils.create_log_folder(config, phase='train')
# writer dict
writer_dict = {
'writer': SummaryWriter(log_dir=output_dict['tb_dir']),
'train_global_steps': 0,
'valid_global_steps': 0,
}
last_epoch = config.TRAIN.BEGIN_EPOCH
if torch.cuda.is_available():
torch.backends.cudnn.benchmark = True
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = get_crnn(config).to(device)
if config.TRAIN.RESUME.IS_RESUME:
model_state_file = config.TRAIN.RESUME.FILE
if model_state_file != '' and os.path.exists(model_state_file):
print('loading pretrained model from %s' % model_state_file)
model.load_state_dict(torch.load(model_state_file))
criterion = torch.nn.CTCLoss(reduction='sum').to(device)
optimizer = optim.Adam(model.parameters(), lr=config.TRAIN.LR)
if isinstance(config.TRAIN.LR_STEP, list):
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer, config.TRAIN.LR_STEP, config.TRAIN.LR_FACTOR, last_epoch - 1)
else:
lr_scheduler = torch.optim.lr_scheduler.StepLR(
optimizer, config.TRAIN.LR_STEP, config.TRAIN.LR_FACTOR, last_epoch - 1)
train_dataset = OcrDataset(config, is_train=True)
train_dataloader = data.DataLoader(
dataset=train_dataset,
batch_size=config.TRAIN.BATCH_SIZE_PER_GPU,
shuffle=config.TRAIN.SHUFFLE,
num_workers=config.WORKERS,
pin_memory=config.PIN_MEMORY,
)
val_dataset = OcrDataset(config, is_train=False)
val_dataloader = data.DataLoader(
dataset=val_dataset,
batch_size=config.TEST.BATCH_SIZE_PER_GPU,
shuffle=config.TEST.SHUFFLE,
num_workers=config.WORKERS,
pin_memory=config.PIN_MEMORY,
)
best_acc = 0.01
converter = utils.strLabelConverter(alphabet)
for epoch in range(last_epoch, config.TRAIN.END_EPOCH):
train(config, train_dataloader, converter, model, criterion, optimizer, device, epoch, writer_dict)
lr_scheduler.step()
acc = validate(config, val_dataloader, converter, model, criterion, device, epoch, writer_dict)
if acc > best_acc:
best_acc = acc
torch.save(model.state_dict(), '{0}/crnn_Rec_done_{1:04d}_{2:.4f}.pth'.format(output_dict['chs_dir'], epoch, acc))
torch.save(model.state_dict(), '{0}/crnn_Rec_best.pth'.format(output_dict['chs_dir']))
writer_dict['writer'].close()