…onTrainer` to optimize loss computation and simplify overrides. (#133)

…UNetTrainer` and added `DiceBCELossWithLogits` as the loss criterion. (#133)

…ted `num_classes` in `UNetTrainer` to 5. (#133)

…resize256`, `roi`) and introduced `UNetSlidingTrainer` for sliding window training. (#133)

…ction naming. (#133)

…ndow` to `validate_case`. (#133)

This reverts commit 0fe8af2.

This reverts commit 9f4008c.

…iding_window` to `validate_case`. (#133)"

This reverts commit b06536d.

…label processing using `Compose` and `build_nnunet_transforms()`. (#133)

…ed compatibility with the specified computation device. (#133)

…etSlidingTrainer` for optimized batch processing. (#133)

…data transfer efficiency to GPU.

…ine recovery and resume workflows.

…er compatibility with `dataclasses`.

…int` for improved clarity and consistency.

…iling models for improved consistency.

… ensure uniform class selection probability.

…s, improving type checks, and refining downsampling logic.

…on preset for type consistency.

…r type consistency.

…nce and adjusted `load` logic to handle preloaded data.

…d updated preload invocation to ensure correct folder usage.

… for better debugging of invalid input types.

…`is_label` flag for improved handling of labeled data.

… handling of label-specific visualizations and colormaps.

…dling of labeled data

… redundant conditions.

… and reorder condition handling in `visualize3d`.

…ors for broader label visualization support.

…em__` methods for improved clarity and maintainability.

… grouping of colors.

…ange access.

…op for cleaner and more efficient training logic.

…t loading.

- Increased min_foreground_samples from 500 to 10000 (matching nnUNet)
- Removed max_foreground_samples parameter (nnUNet has no hard upper limit)
- Fixed sampling logic to match nnUNet: target = max(min(min_samples, total), coverage_requirement)
- Added min_foreground_ratio parameter with foreground ratio check mechanism
- Implemented retry sampling (up to 10 attempts) when force_fg is true but fg_ratio is below threshold

…transformation during dataset preloading.

…ndomized ROI sampling during dataset preparation.

…ng integers are used when loading label images.

… that long integers are used when loading label images."

This reverts commit 4c1128e.