上传文件至 best4_ncnn_model

best4_ncnn_model/model_ncnn.py

@@ -0,0 +1,26 @@
+import numpy as np
+import ncnn
+import torch
+
+def test_inference():
+    torch.manual_seed(0)
+    in0 = torch.rand(1, 3, 640, 640, dtype=torch.float)
+    out = []
+
+    with ncnn.Net() as net:
+        net.load_param("best4_ncnn_model\model.ncnn.param")
+        net.load_model("best4_ncnn_model\model.ncnn.bin")
+
+        with net.create_extractor() as ex:
+            ex.input("in0", ncnn.Mat(in0.squeeze(0).numpy()).clone())
+
+            _, out0 = ex.extract("out0")
+            out.append(torch.from_numpy(np.array(out0)).unsqueeze(0))
+
+    if len(out) == 1:
+        return out[0]
+    else:
+        return tuple(out)
+
+if __name__ == "__main__":
+    print(test_inference())