total_examples = len(dataset)
total_batches = total_examples // batch_size
model = build_model(ishape=ishape, mode='train', net_name=net_name)
# model.summary()
# model.load_weights('{}/weights_.h5'.format(output_path), by_name=True)
min_loss = 2**32
for epoch in range(total_epoches):
# tf.keras.backend.set_value(model.optimizer.learning_rate, 0.001)
gen = genxy(dataset=dataset,
image_dir_path=image_dir_path,
ishape=ishape,
total_batches=total_batches,
batch_size=batch_size)
print('\nTrain epoch {}'.format(epoch))
loss = np.zeros(total_batches)
for batch in range(total_batches):
batchx4d, batchy4d = next(gen)
batch_loss = model.train_on_batch(batchx4d, batchy4d)
loss[batch] = batch_loss
print('-', end='')
if batch % 100 == 99:
print('{:.2f}%'.format((batch + 1) * 100 / total_batches),
end='\n')
# model.load_weights('{}/weights_.h5'.format(output_path), by_name=True)
min_loss = 2**32
max_precision = 0
max_recall = 0
train_dataset = load_dataset(anno_file_path=train_anno_file_path)
test_dataset = load_dataset(anno_file_path=test_anno_file_path)
for epoch in range(total_epoches):
# tf.keras.backend.set_value(model.optimizer.learning_rate, 0.001)
gen = genxy(dataset=train_dataset,
image_dir=train_image_dir,
ishape=ishape,
abox_2dtensor=abox_2dtensor,
iou_thresholds=iou_thresholds,
total_examples=total_train_examples,
total_classes=total_classes,
anchor_sampling=anchor_sampling)
print('\nTrain epoch {}'.format(epoch))
loss = np.zeros(total_train_examples)
for batch in range(total_train_examples):
batchx_4dtensor, batchy_2dtensor, _ = next(gen)
batch_loss = model.train_on_batch(batchx_4dtensor, batchy_2dtensor)
loss[batch] = batch_loss
print('-', end='')
if batch % 100 == 99:
print('{:.2f}%'.format((batch + 1) * 100 / total_train_examples),
abox_2dtensor = tf.reshape(tensor=abox_4dtensor, shape=[-1, 4])
model = build_model(ishape=ishape, resnet_settings=resnet_settings, k=len(asizes), total_classes=total_classes)
# model.summary()
# model.load_weights('{}/weights.h5'.format(output_path), by_name=True)
min_loss = 2**32
max_precision = 0
max_recall = 0
for epoch in range(total_epoches):
gen = genxy(
anno_file_path=train_anno_file_path,
image_dir=train_image_dir,
ishape=ishape,
abox_2dtensor=abox_2dtensor,
iou_thresholds=iou_thresholds,
total_classes=total_classes,
anchor_sampling=anchor_sampling,
mode='train')
print('\nTrain epoch {}'.format(epoch))
loss = np.zeros(total_train_examples)
for batch in range(total_train_examples):
batchx_4dtensor, batchy_2dtensor, _, _ = next(gen)
batch_loss = model.train_on_batch(batchx_4dtensor, batchy_2dtensor)
loss[batch] = batch_loss
if batch%10==9:
print('-', end='')
asizes=asizes[2]),
dtype='float32') # (h3*w3*k3, 4)
a4box_2dtensor = tf.constant(value=genanchors(isize=ishape[:2],
ssize=ssizes[3],
asizes=asizes[3]),
dtype='float32') # (h4*w4*k4, 4)
abox_2dtensors = [
a1box_2dtensor, a2box_2dtensor, a3box_2dtensor, a4box_2dtensor
]
abox_2dtensor = tf.concat(values=abox_2dtensors, axis=0)
dataset = load_dataset(anno_file_path=anno_file_path)
gen = genxy(dataset=dataset,
image_dir=image_dir,
ishape=ishape,
abox_2dtensors=abox_2dtensors,
iou_thresholds=iou_thresholds,
total_examples=total_examples,
total_classes=total_classes,
anchor_samplings=anchor_samplings)
for _ in range(total_examples):
print('{}: 1'.format(datetime.now().time()), end='\n')
batchx_4dtensor, batchy_2dtensor, bboxes = next(gen)
print('{}: 2'.format(datetime.now().time()), end='\n')
pix = batchx_4dtensor[0]
clz_2dtensor = batchy_2dtensor[:, :total_classes +
1] # (h*w*k, total_classes+1)
sum_clz_2dtensor = tf.math.reduce_sum(input_tensor=clz_2dtensor,
axis=-1) # (h*w*k,)