Add half precision setting in README.

README.md

@@ -112,6 +112,7 @@ birefnet = ... # -- BiRefNet should be loaded with codes above, either way.
 torch.set_float32_matmul_precision(['high', 'highest'][0])
 birefnet.to('cuda')
 birefnet.eval()
+birefnet.half()
 
 def extract_object(birefnet, imagepath):
     # Data settings
@@ -123,7 +124,7 @@ def extract_object(birefnet, imagepath):
     ])
 
     image = Image.open(imagepath)
-    input_images = transform_image(image).unsqueeze(0).to('cuda')
+    input_images = transform_image(image).unsqueeze(0).to('cuda').half()
 
     # Prediction
     with torch.no_grad():