STFPM and CFLOW torch inference error

[ke-dev]

Describe the solution you'd like
hi developer, when I use CFLOW and STFPM for torch inference, I encountered the following error, but using PatchCore and PaDim can run normally
the command and error 1
python tools/inference/torch_inference.py \ --config anomalib/models/stfpm/config.yaml \ --weights results/stfpm/mvtec/carpet/weights/model.ckpt \ --input datasets/MVTec/carpet/test/defect_01/ \ --output results/test_img/stfpm/defect_01

Traceback (most recent call last): File "tools/inference/torch_inference.py", line 104, in <module> infer() File "tools/inference/torch_inference.py", line 82, in infer predictions = inferencer.predict(image=image) File "/anomalib/anomalib/anomalib/anomalib/deploy/inferencers/base_inferencer.py", line 99, in predict output = self.post_process(predictions, meta_data=meta_data) File "/anomalib/anomalib/anomalib/anomalib/deploy/inferencers/torch_inferencer.py", line 146, in post_process anomaly_map = predictions.cpu().numpy() RuntimeError: Can't call numpy() on Tensor that requires grad. Use tensor.detach().numpy() instead.

the command and error 2
python tools/inference/torch_inference.py \ --config anomalib/models/cflow/config.yaml \ --weights results/cflow/mvtec/carpet/weights/model.ckpt \ --input datasets/MVTec/carpet/test/defect_01/ \ --output results/test_img/cflow/defect_01

Traceback (most recent call last): File "tools/inference/torch_inference.py", line 104, in <module> infer() File "tools/inference/torch_inference.py", line 82, in infer predictions = inferencer.predict(image=image) File "/anomalib/anomalib/anomalib/anomalib/deploy/inferencers/base_inferencer.py", line 99, in predict output = self.post_process(predictions, meta_data=meta_data) File "/anomalib/anomalib/anomalib/anomalib/deploy/inferencers/torch_inferencer.py", line 174, in post_process anomaly_map, pred_score = self._normalize(anomaly_map, pred_score, meta_data) File "/anomalib/anomalib/anomalib/anomalib/deploy/inferencers/base_inferencer.py", line 166, in _normalize pred_scores = normalize_min_max( File "/anomalib/anomalib/anomalib/anomalib/post_processing/normalization/min_max.py", line 39, in normalize raise ValueError(f"Targets must be either Tensor or Numpy array. Received {type(targets)}") ValueError: Targets must be either Tensor or Numpy array. Received <class 'numpy.float64'>

[samet-akcay]

@ke-dev PR #475 fixes the issue.