#66 Change the file path of the output path to the folder path (#76)

examples/compressor/automatic_compression.py

@@ -10,7 +10,7 @@
 FRAMEWORK = Framework.TENSORFLOW_KERAS
 INPUT_SHAPES = [{"batch": 1, "channel": 3, "dimension": [32, 32]}]
 INPUT_MODEL_PATH = "./examples/sample_models/mobilenetv1.h5"
-OUTPUT_MODEL_PATH = "./outputs/compressed/mobilenetv1_cifar100_automatic.h5"
+OUTPUT_MODEL_PATH = "./outputs/compressed/mobilenetv1_cifar100_automatic"
 COMPRESSION_RATIO = 0.5
 
 compressed_model = compressor.automatic_compression(

examples/compressor/manual_compression.py

@@ -52,7 +52,7 @@
 
 # Compress Model
 COMPRESSED_MODEL_NAME = "test_l2norm"
-OUTPUT_PATH = "./outputs/compressed/graphmodule_manual.pt"
+OUTPUT_PATH = "./outputs/compressed/graphmodule_manual"
 compressed_model = compressor.compress_model(
     compression=compression_1,
     model_name=COMPRESSED_MODEL_NAME,

examples/compressor/recommendation_compression.py

@@ -16,7 +16,7 @@
 TASK = Task.IMAGE_CLASSIFICATION
 FRAMEWORK = Framework.PYTORCH
 INPUT_MODEL_PATH = "./examples/sample_models/graphmodule.pt"
-OUTPUT_MODEL_PATH = "./outputs/compressed/graphmodule_recommend.pt"
+OUTPUT_MODEL_PATH = "./outputs/compressed/graphmodule_recommend"
 INPUT_SHAPES = [{"batch": 1, "channel": 3, "dimension": [224, 224]}]
 COMPRESSION_METHOD = CompressionMethod.PR_L2
 RECOMMENDATION_METHOD = RecommendationMethod.SLAMP

examples/launcher/keras2tflite.py

@@ -1,5 +1,5 @@
 from loguru import logger
-from netspresso.client import SessionClient
+from netspresso.clients.auth import SessionClient
 from netspresso.launcher import (
     ModelConverter,
     ModelBenchmarker,
@@ -14,7 +14,7 @@
     EMAIL = "YOUR_EMAIL"
     PASSWORD = "YOUR_PASSWORD"
     MODEL_PATH = "./examples/sample_models/mobilenetv1.h5"
-    CONVERTED_MODEL_PATH = "./outputs/converted/converted_model.tflite"
+    CONVERTED_MODEL_PATH = "./outputs/converted/keras2tflite"
     session = SessionClient(email=EMAIL, password=PASSWORD)
     converter = ModelConverter(user_session=session)
 
@@ -69,6 +69,7 @@
     benchmarker = ModelBenchmarker(user_session=session)
     benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
         model_path=CONVERTED_MODEL_PATH,
+        target_framework=ModelFramework.TENSORFLOW_LITE,
         target_device_name=DeviceName.RASPBERRY_PI_4B,
     )
     ########################
@@ -77,6 +78,7 @@
     ########################
     # benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
     #     model_path=CONVERTED_MODEL_PATH,
+    #     target_framework=ModelFramework.TENSORFLOW_LITE,
     #     target_device_name=DeviceName.RASPBERRY_PI_4B,
     #     wait_until_done=False,
     # )

examples/launcher/onnx2drpai.py

@@ -1,5 +1,5 @@
 from loguru import logger
-from netspresso.client import SessionClient
+from netspresso.clients.auth import SessionClient
 from netspresso.launcher import ModelConverter, ModelBenchmarker
 from netspresso.launcher import (
     ModelConverter,
@@ -15,7 +15,7 @@
     EMAIL = "YOUR_EMAIL"
     PASSWORD = "YOUR_PASSWORD"
     MODEL_PATH = "./examples/sample_models/test.onnx"
-    CONVERTED_MODEL_PATH = "./outputs/converted/converted_drpai.zip"
+    CONVERTED_MODEL_PATH = "./outputs/converted/onnx2drpai"
     session = SessionClient(email=EMAIL, password=PASSWORD)
     converter = ModelConverter(user_session=session)
 
@@ -64,6 +64,7 @@
     benchmarker = ModelBenchmarker(user_session=session)
     benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
         model_path=CONVERTED_MODEL_PATH,
+        target_framework=ModelFramework.DRPAI,
         target_device_name=DeviceName.RENESAS_RZ_V2L,
     )
     ########################
@@ -72,6 +73,7 @@
     ########################
     # benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
     #     model_path=CONVERTED_MODEL_PATH,
+    #     target_framework=ModelFramework.DRPAI,
     #     target_device_name=DeviceName.RENESAS_RZ_V2L,
     #     wait_until_done=False,
     # )

examples/launcher/onnx2openvino.py

@@ -1,5 +1,5 @@
 from loguru import logger
-from netspresso.client import SessionClient
+from netspresso.clients.auth import SessionClient
 from netspresso.launcher import (
     ModelConverter,
     ModelBenchmarker,
@@ -14,7 +14,7 @@
     EMAIL = "YOUR_EMAIL"
     PASSWORD = "YOUR_PASSWORD"
     MODEL_PATH = "./examples/sample_models/test.onnx"
-    CONVERTED_MODEL_PATH = "./outputs/converted/converted_openvino.zip"
+    CONVERTED_MODEL_PATH = "./outputs/converted/onnx2openvino"
     session = SessionClient(email=EMAIL, password=PASSWORD)
     converter = ModelConverter(user_session=session)
 
@@ -62,6 +62,7 @@
     benchmarker = ModelBenchmarker(user_session=session)
     benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
         model_path=CONVERTED_MODEL_PATH,
+        target_framework=ModelFramework.OPENVINO,
         target_device_name=DeviceName.Intel_XEON_W_2233,
     )
     ########################
@@ -70,6 +71,7 @@
     ########################
     # benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
     #     model_path=CONVERTED_MODEL_PATH,
+    #     target_framework=ModelFramework.OPENVINO,
     #     target_device_name=DeviceName.Intel_XEON_W_2233,
     #     wait_until_done=False,
     # )

examples/launcher/onnx2tensorrt.py

@@ -1,5 +1,5 @@
 from loguru import logger
-from netspresso.client import SessionClient
+from netspresso.clients.auth import SessionClient
 from netspresso.launcher import (
     ModelConverter,
     ModelBenchmarker,
@@ -15,7 +15,7 @@
     EMAIL = "YOUR_EMAIL"
     PASSWORD = "YOUR_PASSWORD"
     MODEL_PATH = "./examples/sample_models/test.onnx"
-    CONVERTED_MODEL_PATH = "./outputs/converted/converted_model.trt"
+    CONVERTED_MODEL_PATH = "./outputs/converted/onnx2tensorrt"
     session = SessionClient(email=EMAIL, password=PASSWORD)
     converter = ModelConverter(user_session=session)
 
@@ -86,6 +86,7 @@
     benchmarker = ModelBenchmarker(user_session=session)
     benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
         model_path=CONVERTED_MODEL_PATH,
+        target_framework=ModelFramework.TENSORRT,
         target_device_name=DeviceName.JETSON_AGX_ORIN,
         target_software_version=SoftwareVersion.JETPACK_5_0_1,
     )
@@ -95,6 +96,7 @@
     ########################
     # benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
     #     model_path=CONVERTED_MODEL_PATH,
+    #     target_framework=ModelFramework.TENSORRT,
     #     target_device_name=DeviceName.JETSON_AGX_ORIN,
     #     target_software_version=SoftwareVersion.JETPACK_5_0_1,
     #     wait_until_done=False,

examples/launcher/onnx2tflite.py

@@ -1,5 +1,5 @@
 from loguru import logger
-from netspresso.client import SessionClient
+from netspresso.clients.auth import SessionClient
 from netspresso.launcher import (
     ModelConverter,
     ModelBenchmarker,
@@ -14,7 +14,7 @@
     EMAIL = "YOUR_EMAIL"
     PASSWORD = "YOUR_PASSWORD"
     MODEL_PATH = "./examples/sample_models/test.onnx"
-    CONVERTED_MODEL_PATH = "./outputs/converted/converted_model.tflite"
+    CONVERTED_MODEL_PATH = "./outputs/converted/onnx2tflite"
     session = SessionClient(email=EMAIL, password=PASSWORD)
     converter = ModelConverter(user_session=session)
 
@@ -73,6 +73,7 @@
 
     benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
         model_path=CONVERTED_MODEL_PATH,
+        target_framework=ModelFramework.TENSORFLOW_LITE,
         target_device_name=DeviceName.RASPBERRY_PI_4B,
     )
     ########################
@@ -81,6 +82,7 @@
     ########################
     # benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
     #     model_path=CONVERTED_MODEL_PATH,
+    #     target_framework=ModelFramework.TENSORFLOW_LITE,
     #     target_device_name=DeviceName.RASPBERRY_PI_4B,
     #     wait_until_done=False,
     # )

examples/launcher/onnx2tflite_int8.py

@@ -1,5 +1,5 @@
 from loguru import logger
-from netspresso.client import SessionClient
+from netspresso.clients.auth import SessionClient
 from netspresso.launcher import (
     ModelConverter,
     ModelBenchmarker,
@@ -16,7 +16,7 @@
     EMAIL = "YOUR_EMAIL"
     PASSWORD = "YOUR_PASSWORD"
     MODEL_PATH = "./examples/sample_models/yolox_auto_compress_0.7.onnx"
-    CONVERTED_MODEL_PATH = "./outputs/converted/converted_model.tflite"
+    CONVERTED_MODEL_PATH = "./outputs/converted/onnx2tflite_int8"
     session = SessionClient(email=EMAIL, password=PASSWORD)
     converter = ModelConverter(user_session=session)
 
@@ -54,6 +54,7 @@
     benchmarker = ModelBenchmarker(user_session=session)
     benchmark_task: BenchmarkTask = benchmarker.benchmark_model(
         model_path=CONVERTED_MODEL_PATH,
+        target_framework=ModelFramework.TENSORFLOW_LITE,
         target_device_name=TARGET_DEVICE_NAME,
         data_type=DATA_TYPE,
         hardware_type=HardwareType.HELIUM,

netspresso/compressor/__init__.py

@@ -1,3 +1,4 @@
+import sys
 from pathlib import Path
 from typing import Dict, List, Union
 from urllib import request
@@ -37,7 +38,7 @@
 )
 from netspresso.enums import ServiceCredit
 
-from ..utils.credit import check_credit_balance
+from ..utils import FileManager, check_credit_balance
 from .utils.onnx import export_onnx
 
 
@@ -254,11 +255,6 @@ def download_model(self, model_id: str, local_path: str) -> None:
             download_link = self.client.get_download_model_link(
                 model_id=model_id, access_token=self.user_session.access_token
             )
-            if not Path(local_path).parent.exists():
-                logger.info(
-                    f"The specified folder does not exist. Local Path: {Path(local_path).parent}"
-                )
-                Path(local_path).parent.mkdir(parents=True, exist_ok=True)
             request.urlretrieve(download_link.url, local_path)
             logger.info(f"Model downloaded at {Path(local_path)}")
 
@@ -441,11 +437,19 @@ def compress_model(
         """
         try:
             logger.info("Compressing model...")
+
+            model_info = self.get_model(compression.original_model_id)
+
+            default_model_path, extension = FileManager.prepare_model_path(
+                folder_path=output_path, framework=model_info.framework
+            )
+
             current_credit = self.user_session.get_credit()
             check_credit_balance(
                 user_credit=current_credit,
                 service_credit=ServiceCredit.ADVANCED_COMPRESSION,
             )
+
             data = CreateCompressionRequest(
                 model_id=compression.original_model_id,
                 model_name=model_name,
@@ -497,12 +501,16 @@ def compress_model(
             self.client.compress_model(
                 data=data, access_token=self.user_session.access_token
             )
+
             self.download_model(
-                model_id=compression_info.new_model_id, local_path=output_path
+                model_id=compression_info.new_model_id,
+                local_path=default_model_path.with_suffix(extension),
             )
             compressed_model = self.get_model(model_id=compression_info.new_model_id)
+
             if compressed_model.framework in [Framework.PYTORCH, Framework.ONNX]:
-                export_onnx(output_path, compressed_model.input_shapes)
+                export_onnx(default_model_path, compressed_model.input_shapes)
+
             logger.info(
                 f"Compress model successfully. Compressed Model ID: {compressed_model.model_id}"
             )
@@ -556,21 +564,19 @@ def recommendation_compression(
 
         try:
             logger.info("Compressing recommendation-based model...")
+
+            default_model_path, extension = FileManager.prepare_model_path(
+                folder_path=output_path, framework=framework
+            )
+
             current_credit = self.user_session.get_credit()
             check_credit_balance(
                 user_credit=current_credit,
                 service_credit=ServiceCredit.ADVANCED_COMPRESSION,
             )
-            model = self.upload_model(
-                model_name=model_name,
-                task=task,
-                framework=framework,
-                file_path=input_path,
-                input_shapes=input_shapes,
-            )
 
             if (
-                model.framework == Framework.PYTORCH
+                framework == Framework.PYTORCH
                 and compression_method == CompressionMethod.PR_NN
             ):
                 raise Exception(
@@ -604,6 +610,14 @@ def recommendation_compression(
                     f"The {compression_method} compression method is only available the VBMF recommendation method."
                 )
 
+            model = self.upload_model(
+                model_name=model_name,
+                task=task,
+                framework=framework,
+                file_path=input_path,
+                input_shapes=input_shapes,
+            )
+
             data = CreateCompressionRequest(
                 model_id=model.model_id,
                 model_name=model_name,
@@ -648,12 +662,16 @@ def recommendation_compression(
             self.client.compress_model(
                 data=data, access_token=self.user_session.access_token
             )
+
             self.download_model(
-                model_id=compression_info.new_model_id, local_path=output_path
+                model_id=compression_info.new_model_id,
+                local_path=default_model_path.with_suffix(extension),
             )
             compressed_model = self.get_model(model_id=compression_info.new_model_id)
+
             if compressed_model.framework in [Framework.PYTORCH, Framework.ONNX]:
-                export_onnx(output_path, compressed_model.input_shapes)
+                export_onnx(default_model_path, compressed_model.input_shapes)
+
             logger.info(
                 f"Recommendation compression successfully. Compressed Model ID: {compressed_model.model_id}"
             )
@@ -699,11 +717,17 @@ def automatic_compression(
 
         try:
             logger.info("Compressing automatic-based model...")
+
+            default_model_path, extension = FileManager.prepare_model_path(
+                folder_path=output_path, framework=framework
+            )
+
             current_credit = self.user_session.get_credit()
             check_credit_balance(
                 user_credit=current_credit,
                 service_credit=ServiceCredit.AUTOMATIC_COMPRESSION,
             )
+
             model = self.upload_model(
                 model_name=model_name,
                 task=task,
@@ -712,22 +736,29 @@ def automatic_compression(
                 input_shapes=input_shapes,
             )
 
+            compressed_model_name = f"{model_name}_automatic_{compression_ratio}"
             data = AutoCompressionRequest(
                 model_id=model.model_id,
-                model_name=f"{model_name}_automatic_{compression_ratio}",
+                model_name=compressed_model_name,
                 recommendation_ratio=compression_ratio,
                 save_path=output_path,
             )
             logger.info("Compressing model...")
             model_info = self.client.auto_compression(
                 data=data, access_token=self.user_session.access_token
             )
-            self.download_model(model_id=model_info.model_id, local_path=output_path)
+
+            self.download_model(
+                model_id=model_info.model_id,
+                local_path=default_model_path.with_suffix(extension),
+            )
             compressed_model = self.model_factory.create_compressed_model(
                 model_info=model_info
-            )
+            )  # TODO: delete
+
             if compressed_model.framework in [Framework.PYTORCH, Framework.ONNX]:
-                export_onnx(output_path, compressed_model.input_shapes)
+                export_onnx(default_model_path, compressed_model.input_shapes)
+
             logger.info(
                 f"Automatic compression successfully. Compressed Model ID: {compressed_model.model_id}"
             )

netspresso/compressor/utils/onnx.py

@@ -31,7 +31,7 @@ def _export_onnx(
 
 def export_onnx(file_path: str, input_shapes: List[int]):
     file_path = Path(file_path)
-    model = torch.load(file_path)
+    model = torch.load(file_path.with_suffix(".pt"))
 
     input_shape = input_shapes[0]
     sample_input = torch.randn((1, input_shape.channel, *input_shape.dimension))