0.4.1 Release Timeline

[calebrob6]

We're planning a 0.4.1 release in approximately a week. This will include all bug fixes in 0.4.1.

If you have any bugs you would like to see fixed, now is the time to submit PRs!

[calebrob6]

TorchGeo 0.4.1 Release Notes

This is a bugfix release. There are no new features or API changes with respect to the 0.4.0 release.

Dependencies

Some dependencies have changed:

• nbmake: 1.3.3+ required now (Faster, more space-efficient tutorials #1124)
• omegaconf: now optional (Deps: make omegaconf/tensorboard optional #1214)
• pytorch-lightning: replaced with lightning (Bump pytorch-lightning[extra] from 1.9.4 to 2.0.0 in /requirements #1178, Fix lightning imports #1179)
• sphinx: 6+ not yet supported (Fix version selection button in docs #1144)
• tensorboard: now optional (Deps: make omegaconf/tensorboard optional #1214)
• pip install torchgeo[all] added, installs all optional dependencies (One extra to rule them all #1095)

Other dependencies now support newer versions:

• black: add 23 support (Bump black[jupyter] from 22.12.0 to 23.1.0 in /requirements #1080)
• kornia: add 0.6.10 support (Bump kornia from 0.6.9 to 0.6.10 in /requirements #1123)
• lightning: add 2 support (Bump pytorch-lightning[extra] from 1.9.4 to 2.0.0 in /requirements #1178)
• mypy: add 1 support (Bump mypy from 0.991 to 1.0.0 in /requirements #1089)
• nbsphinx: add 0.9 support (Bump nbsphinx from 0.8.12 to 0.9.0 in /requirements #1173)
• pandas: add 2 support (Bump pandas from 1.5.3 to 2.0.0 in /requirements #1216)
• pyvista: add 0.38 support (Bump pyvista from 0.37.0 to 0.38.1 in /requirements #1083)
• radiant-mlhub: add 0.5 support (Support for radiant-mlhub 0.5+ #1102)
• scikit-image: add 0.20 support (Bump scikit-image from 0.19.3 to 0.20.0 in /requirements #1153)
• setuptools: add 67 support (Bump setuptools from 66.1.1 to 67.0.0 in /requirements #1066)
• torch: add 2 support (Bump torchvision from 0.14.1 to 0.15.1 in /requirements #1177)
• torchvision: add 0.15 support (Bump torchvision from 0.14.1 to 0.15.1 in /requirements #1177)

Datamodules

• SeCo: fix transforms (Models: fix preprocessing transforms #1166)

Datasets

Fixes for benchmark datasets:

• BigEarthNet: fix order of class labels (Change BigEarthNet class mapping order to match original code #1127)
• CDL: add checksum for 2022 mask (CDL: add 2022 checksum #1201)
• EuroSAT: fix SSL issue, redistribute on Hugging Face (Fix SSL issue with EuroSAT #1065, EuroSAT: redistribute on Hugging Face #1072)
• FAIR1M: fix directory name (FAIR1M Dataset labels_root name #1098, Fix folder name for labels_root in the FAIR1M dataset #1099)
• Landsat: better default bands (Landsat: better default bands #1169)
• UC Merced: redistribute on Hugging Face (UC Merced: redistribute on Hugging Face #1076)
• USAVars: fix class labels (Usa vars label bug #1138)

Fixes for base classes:

• RasterDataset: fix support for datasets where all_bands does not actually contain all bands (e.g., Landsat) (Getting "B1" is not in list error with 0.4.0 release for Landsat8 dataset  #1134, RasterDataset: fix band indexing bug #1135)
• RasterDataset: fix support for datasets where all_bands is not defined and separate_files is False (RasterDataset: fix band indexing bug #1135)
• RasterDataset: fix bug when separate_files and no date in filename_regex (RasterDataset: fix bug when separate_files and no date in regex #1191)
• RasterDataset: remove unnecessary glob (RasterDataset: remove unnecessary glob #1219)
• RasterDataset: better error message when no data found (RasterDataset: better error message when no data found #1193)
• IntersectionDataset: better error message when no overlap (IntersectionDataset: better error message when no overlap #1192)

Models

There are several improvements to our new pre-trained weights:

• Add sha256 suffix for security (Weights: add sha256 suffix #1105)
• Add and improve normalizations (Add seco transforms and zhu normalization to pretrained weights #1119, Models: fix preprocessing transforms #1166)

Trainers

• BYOL: Fix image size to match ViT patch size (Fix load_state_dict for all timm models #1084)
• Fix support for loading ViT weights (utils.load_state_dict() does not support ViT architecture #1049, Fix load_state_dict for all timm models #1084)
• Fix support for non-TensorBoardLogger (Tutorials: issue with trainer and logger? #1143, Trainers: fix support for non-TensorBoardLogger #1145)

Tests

A lot of work in this patch release went towards improving CI:

• Constrain dependencies to avoid CI hang (Constrain dependencies to avoid CI hang #1062)
• Codecov: use repository upload token (Codecov: use repository upload token #1077)
• Cache pip installs (CI: cache pip installs #1057)
• Cancel in-progress jobs on new commit (CI: cancel in-progress jobs when new commit #1094) but not the labeler tasks (labeler: remove concurrency limit #1187)
• Test notebooks when they are modified (CI: test notebooks when modified #1097)
• Speed up object detection tests (Speed up detection tests #1148)
• Fix tests on macOS arm64 (MPS support) (Fix tests when PyTorch built with MPS support #1188)
• Properly test pre-trained model transforms (Models: fix preprocessing transforms #1166)
• Speed up notebook tests (Faster notebook testing #665, Faster, more space-efficient tutorials #1124)

Documentation

• Update the example embedded in the README (README: update Lightning trainer examples #1211)
• Fix broken URLs throughout the documentation (Fix broken URLs #1125)
• Tutorial downloads are now much smaller and faster (Faster, more space-efficient tutorials #1124)
• Replace CSV with TensorBoard in Trainer tutorial (Simplify logging and plotting in Trainers tutorial #1163, Add tensorboard to trainers tutorial #1163 #1189)
• Fix version selection button (Fix version selection button in docs #1144)

Contributors

This release is thanks to the following contributors:

@adamjstewart
@ashnair1
@bugraaldal
@calebrob6
@isaaccorley
@julien-blanchon
@lucastao
@nilsleh
@SpontaneousDuck
@TolgaAktas