Follow ONNX_Runtime_Perf_Tuning and introduce new args

requirements.txt

@@ -12,5 +12,4 @@ tensorflow==2.13.0rc1; sys_platform == 'darwin'
 tensorflow==2.12.0; sys_platform != 'darwin'
 opennsfw2==0.10.2
 protobuf==4.23.2
-tqdm==4.65.0
-threadpoolctl==3.1.0
+tqdm==4.65.0

roop/analyser.py

@@ -1,4 +1,5 @@
 import insightface
+import onnxruntime
 import roop.globals
 
 FACE_ANALYSER = None
@@ -7,6 +8,12 @@ FACE_ANALYSER = None
 def get_face_analyser():
     global FACE_ANALYSER
     if FACE_ANALYSER is None:
+        session_options = onnxruntime.SessionOptions()
+        if roop.globals.gpu_vendor is not None:
+            session_options.intra_op_num_threads = roop.globals.gpu_threads
+        else:
+            session_options.intra_op_num_threads = roop.globals.cpu_threads
+        session_options.execution_mode = onnxruntime.ExecutionMode.ORT_PARALLEL
         FACE_ANALYSER = insightface.app.FaceAnalysis(name='buffalo_l', providers=roop.globals.providers)
         FACE_ANALYSER.prepare(ctx_id=0, det_size=(640, 640))
     return FACE_ANALYSER

roop/core.py

@@ -101,17 +101,6 @@ def pre_check():
         roop.globals.providers = ['CPUExecutionProvider']
 
 
-def start_processing():
-    # gpu mode
-    if roop.globals.gpu_vendor is not None:
-        process_video(args['source_img'], args["frame_paths"])
-        return
-    # cpu mode
-    with threadpool_limits(limits=roop.globals.cpu_threads):
-        process_video(args['source_img'], args["frame_paths"])
-        return
-
-
 def preview_image(image_path):
     img = Image.open(image_path)
     img = img.resize((180, 180), Image.ANTIALIAS)
@@ -223,7 +212,7 @@ def start():
         key=lambda x: int(x.split(sep)[-1].replace(".png", ""))
     ))
     status("swapping in progress...")
-    start_processing()
+    process_video(args['source_img'], args["frame_paths"])
     status("creating video...")
     create_video(video_name, exact_fps, output_dir)
     status("adding audio...")

roop/swapper.py

@@ -15,10 +15,9 @@ def get_face_swapper():
         session_options = onnxruntime.SessionOptions()
         if roop.globals.gpu_vendor is not None:
             session_options.intra_op_num_threads = roop.globals.gpu_threads
-            session_options.execution_mode = onnxruntime.ExecutionMode.ORT_PARALLEL
         else:
-            session_options.enable_cpu_mem_arena = True
-        session_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
+            session_options.intra_op_num_threads = roop.globals.cpu_threads
+        session_options.execution_mode = onnxruntime.ExecutionMode.ORT_PARALLEL
         model_path = os.path.join(os.path.abspath(os.path.dirname(__file__)), '../inswapper_128.onnx')
         FACE_SWAPPER = insightface.model_zoo.get_model(model_path, providers=roop.globals.providers, session_options=session_options)
     return FACE_SWAPPER