def __call__(self, x, y, norm):
x = self.generator(x)
loss = self.criterion(x.contiguous().view(-1, x.size(-1)),
y.contiguous().view(-1)) / norm
loss.backward()
if self.opt is not None:
self.opt.step()
self.opt.optimizer.zero_grad()
return loss.data * norm
# Train the simple copy task.
V = 11
criterion = LabelSmoothing(size=V, padding_idx=0, smoothing=0.0)
model = make_model(V, V, N=2)
model_opt = NoamOpt(
model.src_embed[0].d_model, 1, 400,
torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))
for epoch in range(10):
model.train()
run_epoch(data_gen(V, 30, 20), model,
SimpleLossCompute(model.generator, criterion, model_opt))
model.eval()
print(
run_epoch(data_gen(V, 30, 5), model,
SimpleLossCompute(model.generator, criterion, None)))
def greedy_decode(model, src, src_mask, max_len, start_symbol):
def __init__(self): self._model = train.make_model() self._model.compile(loss='mean_squared_error', optimizer='adadelta')
def __init__(self): self._model = train.make_model() self._model.compile(loss='mean_squared_error', optimizer='adadelta')
#Load Model and Generate Predictions
import tensorflow as tf
from decode_predictions import DecodePredictions
from train import make_model
from utils import resize_and_pad_image
from dataset import data
import tensorflow_datasets as tfds
from utils import visualize_detections
# Change this to `model_dir="retinanet/"` when not using the downloaded weights
weights_dir = "data"
model = make_model()
latest_checkpoint = tf.train.latest_checkpoint(weights_dir)
model.load_weights(latest_checkpoint)
image = tf.keras.Input(shape=[None, None, 3], name="image")
predictions = model(image, training=False)
detections = DecodePredictions(confidence_threshold=0.5)(image, predictions)
inference_model = tf.keras.Model(inputs=image, outputs=detections)
def prepare_image(image):
image, _, ratio = resize_and_pad_image(image, jitter=None)
image = tf.keras.applications.resnet.preprocess_input(image)
return tf.expand_dims(image, axis=0), ratio
_, _, dataset_info = data()
])
img_resize = resize(img)
img_for_pred = norm(img_resize).unsqueeze(dim=0)
attns = []
start = time.time()
text = predict(model, pe, img_for_pred, max_len, attns=attns)[0]
end = time.time()
text = text[:text.find('[s]')]
print(f"time consumed:{time_interval(end - start)}")
for i, char in enumerate(text):
if torch.cuda.is_available():
attn = attns[i].cpu()
attn = attn.detach().numpy()
show(img_resize, char, attn)
show(img_resize, text)
if __name__ == '__main__':
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model, pe = make_model()
model.load_state_dict(
torch.load(f"./model_params_{2.0}.pkl", map_location="cpu"))
if torch.cuda.is_available():
model = model.cuda()
pe = pe.cuda()
image = cv2.imread(
"/data1/luofuyou/ReCTS_test/test_ReCTS_task2_000229.jpg")
image = Image.fromarray(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
visualize(model, pe, image)