def setUpClass(cls):
export_config = get_config(action='test', optimised=True)
cls.config = export_config
if not os.path.exists(cls.config['checkpoint']):
download_checkpoint()
if os.path.exists(export_config["default_image_path"]):
cls.image_path = export_config["default_image_path"]
else:
if not os.path.exists(export_config["image_path"]):
download_data()
cls.image_path = export_config["image_path"]
dataset_test = RSNADataSet(cls.config['dummy_valid_list'],
cls.config['dummy_labels'],
cls.image_path,
transform=True)
cls.data_loader_test = DataLoader(dataset=dataset_test,
batch_size=1,
shuffle=False,
num_workers=4,
pin_memory=False)
alpha = cls.config['alpha']**cls.config['phi']
beta = cls.config['beta']**cls.config['phi']
cls.model = DenseNet121Eff(alpha, beta, cls.config['class_count'])
cls.class_names = cls.config['class_names']
cls.device = torch.device(
"cuda" if torch.cuda.is_available() else "cpu")
cls.inference = RSNAInference(cls.model, cls.data_loader_test,
cls.config['class_count'],
cls.config['checkpoint'],
cls.config['class_names'],
cls.device)
cls.exporter = Exporter(cls.config, optimised=True)
def test_export_onnx(self):
self.config = get_config(action='export', optimised=True)
if not os.path.exists(self.config['checkpoint']):
download_checkpoint()
self.exporter = Exporter(self.config, optimised=True)
self.exporter.export_model_onnx()
checkpoint = os.path.split(self.config['checkpoint'])[0]
self.assertTrue(
os.path.join(checkpoint, self.config.get('model_name_onnx')))
def test_trainer(self):
self.model = DenseNet121(self.config["clscount"])
if not os.path.exists(self.config["checkpoint"]):
download_checkpoint()
self.device = self.config["device"]
self.trainer = RSNATrainer(
self.model, self.data_loader_train, self.data_loader_valid,
self.data_loader_test, self.config["clscount"],
self.config["checkpoint"], self.device,
self.config["class_names"], self.config["lr"])
self.trainer.train(self.config["max_epoch"],
self.config["savepath"])
cur_train_loss = self.trainer.current_train_loss
self.trainer.train(self.config["max_epoch"],
self.config["savepath"])
self.assertLessEqual(self.trainer.current_train_loss,
cur_train_loss)
def test_export_ir(self):
self.config = get_config(action='export', optimised=True)
if not os.path.exists(self.config['checkpoint']):
download_checkpoint()
self.exporter = Exporter(self.config, optimised=True)
self.model_path = os.path.split(self.config['checkpoint'])[0]
if not os.path.exists(
os.path.join(self.model_path,
self.config.get('model_name_onnx'))):
self.exporter.export_model_onnx()
self.exporter.export_model_ir()
name_xml = self.config.get('model_name') + '.xml'
name_bin = self.config.get('model_name') + '.bin'
xml_status = os.path.exists(os.path.join(self.model_path,
name_xml))
bin_status = os.path.exists(os.path.join(self.model_path,
name_bin))
self.assertTrue(xml_status)
self.assertTrue(bin_status)
def setUpClass(cls):
cls.config = get_config(action='export')
if not os.path.exists(cls.config['checkpoint']):
download_checkpoint()
cls.model_path = cls.config['checkpoint']