Beat nnU-Net#168

Draft

ATATC wants to merge 318 commits intomainfrom

This pull request is big! We're only showing the most recent 250 commits

Commits on Dec 28, 2025

Added `frontend` parameter to benchmark functions for improved configurability. (#133)
ATATC
committed
Added support for `NotionFrontend` and `WandBFrontend` in the benchmark script's `frontend` parameter. (#133)
ATATC
committed
Simplified `frontend` parameter lookup in `benchmark.py`. (#133)
ATATC
committed

Commits on Dec 29, 2025

Adjusted ROI shape and sliding window configurations, and updated resize functions to include `frontend` parameter. (#133)
ATATC
committed
Revert "Temporary fix 2."
ATATC
committed
Revert "Temporary optimization."
ATATC
committed
Removed sliding window and related functionality. (#192)
ATATC
committed
Reintroduced sliding window functionality with `do_sliding_window()`, `revert_sliding_window()`, and `slide_dataset()` methods. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Extended sliding window module with `UnsupervisedSWDataset` and `SupervisedSWDataset` classes, and updated related methods. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Refactored `benchmark.py` to enhance dataset handling with `sliding_window` and new `SupervisedSWDataset`. Removed unused `roi` function. Improved device and dataset initialization logic. (#133)
ATATC
committed
Removed `UNetSlidingTrainer` and replaced its usage with `UNetTrainer` in `benchmark.py`. Cleaned up unused imports and methods in `unet.py`. (#133)
ATATC
committed
Removed `SlidingPredictor` and associated sliding window functionality. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Adjusted validation dataset path to account for sliding window preprocessing in `benchmark.py`. (#133)
ATATC
committed
Refactored shape handling in sliding window function to improve dimension compatibility. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Refactored `do_sliding_window` to streamline padding logic with `Pad2d` and `Pad3d`, enhancing 2D/3D shape handling. (#192)
ATATC
committed
Refactored `do_sliding_window` to streamline padding logic with `Pad2d` and `Pad3d`, enhancing 2D/3D shape handling. (#192)
ATATC
committed
Updated `do_sliding_window` to compute `shape` from `window_shape` and `overlap`, ensuring consistent dimensions. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Adjusted `do_sliding_window` to use `stride` instead of `shape`, refining 2D/3D window padding and unfolding logic. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Replaced division-based index calculation with `log10` for consistent zero-padding in sliding window filenames. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Removed `device` argument from `torch.load` in `load` method for compatibility with default behavior. (#192)
ATATC
committed
Merge branch '192' into 133
ATATC
committed
Integrated progress tracking with `rich` in sliding window operations, enhancing dataset processing visibility. (#192)
ATATC
committed
Added `fast_save` and `fast_load` utilities, integrated tensor preloading in datasets, and enhanced sliding window handling with `TensorLoader`. (#192)
ATATC
committed
Increased batch size and updated sliding window shape for validation dataset preparation in `benchmark.py`. (#133)
ATATC
committed
Merge remote-tracking branch 'origin/192' into 133
ATATC
committed
Added the SlidingTrainer class to support sliding window training and updated the import statement to include the relevant functionality. (#192)
perctrix
committed
Padded label to match reconstructed shape in SlidingTrainer validation. (#192)
perctrix
committed

Commits on Dec 30, 2025

Commits on Dec 31, 2025

Commits on Jan 2, 2026

Refactored SlidingTrainer to use dataset methods instead of hardcoded loading. (#192)
perctrix
committed

Commits on Jan 3, 2026

Commits on Jan 6, 2026

Commits on Jan 7, 2026

Commits on Jan 8, 2026

Commits on Jan 9, 2026

Commits on Jan 10, 2026

Commits on Jan 11, 2026

Commits on Jan 12, 2026

Commits on Jan 13, 2026

Updated nnUNet style transform
perctrix
committed

Commits on Jan 14, 2026

Commits on Jan 15, 2026

Commits on Jan 16, 2026

Commits on Jan 19, 2026

Aligned RandomROIDataset with nnUNet foreground sampling strategy.
perctrix
committed

Commits on Jan 22, 2026

Commits on Jan 24, 2026

Commits on Jan 25, 2026

Commits on Jan 26, 2026

Refactored `construct_new` method in `inspection.py` to initialize `images` and `labels` for enhanced object state consistency.
ATATC
committed

Commits on Jan 27, 2026

Commits on Jan 29, 2026

Commits on Jan 30, 2026

Clipped gradients in segmentation preset to prevent exploding gradients; removed unused model compilation logic in training setup.
ATATC
committed

Commits on Jan 31, 2026

Added debug image saving at epoch 80 in benchmark UNetTrainer to diagnose low dice scores.
perctrix
committed

Commits on Feb 1, 2026

Revert "Added debug image saving at epoch 80 in benchmark UNetTrainer to diagnose low dice scores."
ATATC
committed

Commits on Feb 2, 2026

Commits on Feb 3, 2026

Refined loss initialization in segmentation preset to ensure device and dtype consistency.
ATATC
committed

Commits on Feb 4, 2026