ultralytics 8.0.122 Fix torch.Tensor inference (#3363)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: krzysztof.gonia <4281421+kgonia@users.noreply.github.com>

tests/test_python.py

@@ -6,6 +6,7 @@ import cv2
 import numpy as np
 import torch
 from PIL import Image
+from torchvision.transforms import ToTensor
 
 from ultralytics import RTDETR, YOLO
 from ultralytics.yolo.data.build import load_inference_source
@@ -70,7 +71,7 @@ def test_predict_img():
     # Test tensor inference
     im = cv2.imread(str(SOURCE))  # OpenCV
     t = cv2.resize(im, (32, 32))
-    t = torch.from_numpy(t.transpose((2, 0, 1)))
+    t = ToTensor()(t)
     t = torch.stack([t, t, t, t])
     results = model(t, visualize=True)
     assert len(results) == t.shape[0]