Releases: wolny/pytorch-3dunet
Releases · wolny/pytorch-3dunet
Random patch scaling, performance improvements
What's Changed
- fix overwriting of prediction for multiple thresholds by @Josh-Talks in #115
- Fix/missing return statement by @shota-0129 in #118
- checkpoint_dir is a required argument by @sai8951 in #120
- Refactor and improvements by @wolny in #121
New Contributors
- @Josh-Talks made their first contribution in #115
- @shota-0129 made their first contribution in #118
- @sai8951 made their first contribution in #120
Full Changelog: 1.8.7...1.9.0
Allow channel-wise normalization, prevent stride_shape override in prediction
1.8.7 Bump version: 1.8.6 → 1.8.7
Fix tiling artifacts during prediction
Add 'save_segmentation' option to prediction
See #112 for details
Fix hdf5 lazy loader
1.8.4 Bump version: 1.8.3 → 1.8.4
Pixel-wise weights training improvements
- improve training with
PixelWiseCrossEntropyLoss
- add examples for training with
PixelWiseCrossEntropyLoss
Improve ResidualUNet
Allow arbitrary patch sizes to be passed to ResidualUNet
Bug fix
Bug fix
Additional features and code improvements
- Override config parameters from command-line in prediction
- Move tensorboard files to separate subfolders in CHECKPOINT_DIR
- Allow other optimizers to be used apart from Adam