TRT `--half` fix autocast images to FP16 (#8435)

* TRT `--half` fix autocast images to FP16

Resolves bug raised in https://github.com/ultralytics/yolov5/issues/7822

* Update common.py

models/common.py

@@ -441,6 +441,9 @@ class DetectMultiBackend(nn.Module):
     def forward(self, im, augment=False, visualize=False, val=False):
         # YOLOv5 MultiBackend inference
         b, ch, h, w = im.shape  # batch, channel, height, width
+        if self.fp16 and im.dtype != torch.float16:
+            im = im.half()  # to FP16
+
         if self.pt:  # PyTorch
             y = self.model(im, augment=augment, visualize=visualize)[0]
         elif self.jit:  # TorchScript