How to inference custom dataset without Folder()?

[cory619-seungwon]

Hello all.

I've successfully trained my custom dataset!
However, I'm struggling to inference my test dataset.

I'd like to inference test dataset all but, when I'm using Folder() to inference test dataset as below, it leads to split my dataset.
datamodule = Folder( name='evaluation', root=str(evaluation_dir), normal_dir=str(evaluation_dir), task=TaskType.CLASSIFICATION, eval_batch_size=1, )

I don't wanna split my test. then, I'd like to inference my dataset all.
Could you all help me?

Any method inference only 1 image?
I think, in order to inference only 1 image, Folder() should be deleted... right?
I don't understand why Folder() must have split ratio 0~1...

[AlexandreFyl]

Hello,

You can use TorchInferencer to achieve this, as follows :

inferencer = TorchInferencer(path=model_path)
result= inferencer.predict(img_path)

Your score will be available in the pred_score key !

[cory619-seungwon]

Thanks a lot. I got a solution as below.

    inferencer = TorchInferencer(path="D:\\patchcore\\anomalib_2\\models\\model.pt", device="cpu")

    image = Image.open("D:\\patchcore\\anomalib_2\\evaluation_1_image\\TEST_098.png").convert("RGB")
    image = image.resize((224, 224))
    transform = transforms.ToTensor()
    image_tensor = transform(image) 
    
    result = inferencer.predict(image=image_tensor)

[samet-akcay]

@cory619-seungwon, you may not even need the transforms, which are already within the model graph for the exported models. Just pass your input image, and the model should apply the transforms under the hood. You could refer to the following link for details.
https://anomalib.readthedocs.io/en/v1.1.1/markdown/guides/how_to/data/transforms.html

[cory619-seungwon]

Let me close my query. Many thanks to all.