update

test_script/retinaface_pure_impl.py

@@ -34,6 +34,8 @@ torch.use_deterministic_algorithms = True
 image = Image.open(r"C:\Users\tomokazu\CLionProjects\ameba_blog_downloader\manaka_test.jpg").convert(mode="RGB")
 image_arr = from_numpy(np.array(object=image, dtype=np.float32)).unsqueeze(0).permute(0, 3, 1, 2)
 
+py_model: Model = get_model(model_name='resnet50_2020-07-20', max_size=512)
+print(py_model.predict_jsons(array(image)))
 max_size = 512
 
 example_input = randn(size=[10, 3, 256, 256]).float()

test_script/to_onnx.py

@@ -1,27 +1,33 @@
 from torch import load, randn, float, half, jit, ones, no_grad
-import torch_tensorrt
+# import torch_tensorrt
+from torchinfo import summary
 from torch.nn import Module
 from torch.onnx import export
 
 model: Module = load(
-    f='/home/tomokazu/PycharmProjects/helloproject-ai/data/artifact/facenet-tl_2023-10-15 14:46:51.187699/checkpoints/80.pth')
-model.cuda()
+    f=r"\\tomokazu-ubuntu-server\share\helloproject-ai-data\artifact\facenet-tl_2023-10-22 213825.539264\model.pth")
+# model.cuda()
 model.eval()
-model = model.half()
+summary(
+    model=model,
+    input_size=[1, 3, 224, 224],
+    device='cpu',
+    col_names=["input_size", "output_size", "num_params", "params_percent", "kernel_size", "mult_adds", "trainable"]
+)
 with no_grad():
-    example_input = randn(1, 3, 224, 224).cuda().half()
+    example_input = randn(1, 3, 224, 224)
 
     export(
         model=model,
         args=example_input,
-        f="onnx_test.onnx",
+        f="face_recognition.onnx",
         input_names=["input"],
         output_names=["output"],
         dynamic_axes={
             "input": {
                 0: "batch_size",
-                2: "height",
-                3: "width"
+                # 2: "height",
+                # 3: "width"
             }
         },
         verbose=False