Pvinspect

New

* `save_images` now has an additional argument `filename_hook` that allows to modify the generation of image filenames
* Meta data is now cached to speed up loading of long sequences
* `ClassificationDataset.result_sequence` now accepts results of type `torch.float` as well

Fixed

* `PartialModuleImage.cols` and `PartialModuleImage.rows` is now set correctly with `segment_module_part` and negative `first_col`/`first_row`
* Fix semantic of `first_row` `first_col` in combination with module segmentation

New

* Add support for uint8 images

Fixed

* Fix occasionally failing checkerboard detection

Fixed/Improved

* Add missing dependency

New

* Add performance test for single module detection
* Pandas methods are now (all) available from `ImageSequence.pandas`
* `Image` now supports lazy loading of data (which is now default)
* Introduced convenience method `from_self` to avoid using `type(self).from_other(self, ...)`
* Add PyTorch `Dataset` and `ClassificationDataset` implementation
* Add ELPV cell image dataset

Fixed/Improved

* Speed up meta data handling

* Prefix/suffix for filenames in `save_images`

* Improve robustness of reference detection
* Fix plot with visualization for more than one image
* Add `ImageSequence.drop_duplicates`
* Remove unused `copy` argument to `CellImageSequence`
* Store reference to original for `CellImage` with the correct cell marked
* Avoid passing geometric entities to segmented modules
* Save image meta in separate JSON file
* Clip image after FF calibration if it exceeds datatype limits and issue a warning
* Refactor modality such that it is an Enum
* All Image attributes (except data) are meta data now