0.5.1 Release Timeline

[nilsleh]

We're planning a 0.5.1 release next week. This will include all bug fixes in 0.5.1.

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

[nilsleh]

This is a bugfix release. There are no new features or API changes with respect to the 0.5.0 release.

Datamodules

• EuroSAT: make channel normalization statistics responsive to dynamic band selection (EuroSATDataModule initialised with 3 bands gives error: mean length and number of channels do not match #1634, EuroSATDataModule: set mean/std based on bands #1681)

Datasets

• AGB Live Woody Biomass: update download link for dataset (HTTP internal server error when trying to download AbovegroundLiveWoodyBiomassDensity #1679, Fix Download Link AGB Live Woody Biomass dataset #1713)
• EuroSAT: remove classes attribute and instead rely on ImageFolder classes (Bug in EuroSAT.plot() #1648, Fix classes in EuroSAT #1650)
• OSCD: change image datatype be float instead of int (OSCD RuntimeError: Input type (torch.cuda.LongTensor) and weight type (torch.cuda.FloatTensor) should be the same #1652, Format OSCD images to float #1656)
• RESICS45: remove classes attribute and instead rely on ImageFolder classes (Bug in EuroSAT.plot() #1648, Fix classes in EuroSAT #1650)
• UC Merced: fix plotting which expects images from dataset to be normalized already (Fix for Blank Images in plot due to Float Tensor Ranges #1712)
• UC Merced: remove classes attribute and instead rely on ImageFolder classes (Bug in EuroSAT.plot() #1648, Fix classes in EuroSAT #1650)
• GeoDataset: check if the path points to a Virtual File System, to prevent error of looking and not finding the paths locally (Download support broken for some GeoDatasets if directory does not exist #1605, Add check if path is vsi #1612)
• GeoDatasets: consistent use of paths argument instead of root in RuntimeError of several datasets (Fixed path and root inconsistency. #1704, Fix another path/root inconsistency #1717)

Trainers

• During logging, trainers were expecting a datamodule with plot functionality, which was preventing trainers from being used with custom Pytorch Dataloaders (Fix all Trainers to support PyTorch DataLoaders #1703)
• Remove default callback configurations of trainers and leave it to user instead (trainer.test results not reproduced when loading from a checkpoint #1640, Unable to log_model="all" with wandb_logger #1641, trainers docs show configuring callbacks but these will not be used #1642, Checkpoint saving not working as expected #1645, Remove configure_callbacks #1647)
• Skip weights and augmentations when saving hparams, allowing these parameters to be changed (Modifying MoCo augmentations breaks training #1622, load_from_checkpoint ValueError.. is not a valid ResNet18_Weights #1639, Trainers: skip weights and augmentations when saving hparams #1670)

Scripts

• Solve logging conflict by allowing config.yaml file to be overwritten (Wandb logger conflict using LightningCLI #1621, Pass save_config_kwargs to LightningCLI #1625)

Tests

• Greatly reduce memory footprint of CI which was causing PR tests to fail (CI: minimize storage usage #1658)
• Copy testing csv file instead of downloading it for MapInWild dataset test (MapInWild: prevent download during testing #1657)
• Fix choco install unrar in CI by using 7zip instead of unrar (CI: fix RAR backend installation #1697)
• CI: use unique names for release caches (CI: use unique names for release caches #1723)

Documentation

• README: update SemanticSegmentationTask example with arguments introduced in 0.5 (SemanticSegmentationTask Signature in README #1608)
• README: add section on LightningCLI usage with torchgeo (Error in main readme: we provide train.py and evaluate.py scripts to train and evaluate #1626, Update info on LightningCLI #1628)
• README: add section on availability of pretrained weights in torchgeo (Update README with paragraph about available pretrained weights #1716)
• BioMassters: fix typo in docs' overview table of non-geo datasets (Fix formatting for BioMassters #1718)
• SSL4EO-L Benchmark: add dataset information to documentation (Add SSL4EO-L Benchmark to docs #1719)

Contributors

This release is thanks to the following contributors (in alphabetical order):

@adamjstewart
@ashnair1
@dylanrstewart
@kaybe20
@menglutao
@nilsleh
@pioneerHitesh
@robmarkcole