[Bug]: Use of Tiler.untile not possible without running tile in the same scope

[AntiLibrary5]

Describe the bug

Using untile() without running tile() first is not possible. Also this is not clear in the documentation

Dataset

N/A

Model

N/A

Steps to reproduce the behavior

Following gives an idea when tiling has been done separately and it is in interest to untile later:

    inference_dataset = WaferDataset(config, stage='inference')
    inference_dataset.dataset.load_image_paths(inference_paths)
    inference_loader = inference_dataset.get_dataloader()
    for tiles, metadata in inference_loader:
        logger.info(tiles.shape)
        logger.info(metadata)
        # untile
        untiled = inference_dataset.dataset.untile_image(tiles=tiles.squeeze(0), metadata=metadata)
        logger.info(untiled.shape)

OS information

OS information:

OS: Win11
Python version: 3.10
Anomalib version: 1.2
PyTorch version: 2.5

Expected behavior

Hi
As the title mentions, it should be possible to use the untile() given the right params. But the way the current implementation is, one needs to run tile() on an input tensor which initializes a bunch of variables for the object of the Tiler which are used downstream for untiling. So if someone wants to tile at one point and then untile at another (imagine a usecase of saving tiled images then loading them up and untiling separately), it wont work

Screenshots

No response

Pip/GitHub

pip

What version/branch did you use?

No response

Configuration YAML

NA

Logs

File "C:\Users\v.arora\PycharmProjects\IterMaskUnity\src\data\infer\dataset.py", line 55, in untile_image
    return self.tiler.untile(tiles)
  File "C:\Users\v.arora\AppData\Local\anaconda3\envs\itermask_unity\lib\site-packages\anomalib\data\utils\tiler.py", line 431, in untile
    image = self.__fold(tiles)
  File "C:\Users\v.arora\AppData\Local\anaconda3\envs\itermask_unity\lib\site-packages\anomalib\data\utils\tiler.py", line 302, in __fold
    image_size = (self.batch_size, num_channels, int(self.resized_h * scale_h), int(self.resized_w * scale_w))
AttributeError: 'Tiler' object has no attribute 'batch_size'

Code of Conduct

• I agree to follow this project's Code of Conduct

[blaz-r]

Hi, thanks for reporting this. Tiler is intended to be used within the model, and is then exported together with model, preserving the batch size attribute. I had a similar issue when working on the tiled ensemble and fixed it by introducing this code before calling tiler untile:
https://github.com/blaz-r/anomalib/blob/8e3d9d501da9db2aa754f3f0463718e2985af4e7/src/anomalib/pipelines/tiled_ensemble/components/utils/ensemble_tiling.py#L96-L100
this will not immediately work here, since the default shapes are different, but the idea should apply. Basically, when calling untile, you can just set the batch size based on the current input. Maybe this could be added to the base Tiler class, to enable untiling across different instances.
Hope that helps.