def test_reload_model(test_dataset, test_anchors, test_masks, test_classes):
img_shape = test_dataset.target_shape
model = YoloV3(img_shape,
test_dataset.max_objects,
backbone='MobileNetV2',
anchors=test_anchors,
num_classes=len(test_classes),
training=True)
loss_fn = losses.make_loss(model.num_classes, test_anchors, test_masks,
img_shape[0], len(test_dataset))
optimizer = model.get_optimizer('adam', 1e-4)
model.compile(optimizer, loss_fn, run_eagerly=True)
model.fit(test_dataset, test_dataset, 1)
save_path = BASE_PATH.parent / 'model.h5'
model.save(save_path)
del model
model = YoloV3(img_shape,
test_dataset.max_objects,
backbone='MobileNetV2',
anchors=test_anchors,
num_classes=len(test_classes),
training=False)
model.load_weights(save_path)
_, _, _, valid_detections = model.predict(tf.zeros((1, *img_shape)))
assert len(valid_detections) > 0
def test_model_resnet50(test_dataset, test_anchors, test_masks, test_classes):
img_shape = test_dataset.target_shape
model = YoloV3(img_shape,
test_dataset.max_objects,
backbone='ResNet50V2',
anchors=test_anchors,
num_classes=len(test_classes),
training=True)
loss_fn = losses.make_loss(model.num_classes, test_anchors, test_masks,
img_shape[0], len(test_dataset))
optimizer = model.get_optimizer('sgd', 1e-4)
model.compile(optimizer, loss_fn, run_eagerly=True)
res = model(test_dataset[0][0])
assert res is not None
def test_model_DenseNet121_grid608_64(test_dataset_grid608_64, test_anchors,
test_masks, test_classes):
img_shape = test_dataset_grid608_64.target_shape
model = YoloV3(img_shape,
test_dataset_grid608_64.max_objects,
backbone='DenseNet121',
anchors=test_anchors,
num_classes=len(test_classes),
training=True)
loss_fn = losses.make_loss(model.num_classes, test_anchors, test_masks,
img_shape[0], len(test_dataset_grid608_64))
optimizer = model.get_optimizer('sgd', 1e-4)
model.compile(optimizer, loss_fn, run_eagerly=True)
res = model(test_dataset_grid608_64[0][0])
assert res is not None