if gpus:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
anchor_sizes = tf.convert_to_tensor(l_config.anchor_sizes, tf.float32)
anchor_sizes = anchor_sizes / l_config.image_target_size[0]
anchor_instance = l_anchors.Anchor(anchor_sizes, l_config.grid_sizes,
l_config.image_target_size)
anchors = anchor_instance.get_anchors()
train_parse = l_datasets.Parse(l_config.train_image_dir, anchors,
l_config.grid_sizes, l_config.image_target_size)
val_parse = l_datasets.Parse(l_config.val_image_dir, anchors,
l_config.grid_sizes, l_config.image_target_size)
model = l_models.YoloV3(l_config.filters, anchors, l_config.grid_sizes,
l_config.class_num)
last_time = time.time()
count = 0
images = tf.ones([1] + list(l_config.image_target_size) + [3])
@tf.function
def pre(images):
return model(images, training=False)
while True:
now = time.time()
if now - last_time > 1:
print(now - last_time)
val_parse = l_datasets.Parse(l_config.val_image_dir, anchors,
l_config.grid_sizes, l_config.image_target_size)
train_ds = tf.data.TextLineDataset(l_config.train_label_file)
train_ds = train_ds.map(train_parse)
train_ds = train_ds.shuffle(128)
train_ds = train_ds.batch(l_config.batch_size)
train_ds = train_ds.prefetch(tf.data.experimental.AUTOTUNE)
val_ds = tf.data.TextLineDataset(l_config.val_label_file)
val_ds = val_ds.map(val_parse)
val_ds = val_ds.batch(l_config.batch_size)
val_ds = val_ds.prefetch(tf.data.experimental.AUTOTUNE)
model = l_models.YoloV3(anchors,
l_config.grid_sizes,
l_config.class_num,
is_decode=True)
loca_loss = l_losses.LocationLoss(anchors)
conf_loss = l_losses.ConfidenceLoss()
cate_loss = l_losses.CategoricalLoss()
all_loss = l_losses.AllLoss()
loca_metric = l_metrics.Location()
conf_metric = l_metrics.Confidence()
true_conf_metric = l_metrics.TrueConfidence()
false_conf_metric = l_metrics.FalseConfidence()
cate_metric = l_metrics.Categorical()
precision = l_metrics.Precision()
recall = l_metrics.Recall()