v2.0

What's Changed

• New pretrained architectures: ConvNeXT, EfficientNetV2, MaxViT, and more. Each in several sizes. Available on HuggingFace.
• Reworked finetuning procedure. All these architectures are finetuneable through a common method.
• Reworked finetuning options. Batch norm finetuning removed. Cosine schedule option added.
• Reworked finetuning saving/loading. Auto-downloads encoder from HuggingFace.
• Now supports regression finetuning (as well as multi-class and binary). See pytorch/examples/finetuning
• Updated timm to 0.9.10, allowing latest model architectures. Previously downloaded checkpoints may not load correctly!
• (internal until published) GZ Evo v2 now includes Cosmic Dawn (HSC H2O). Significant performance improvement on HSC finetuning. Also now includes GZ UKIDSS (dragged from our archives).
• Updated pytorch to 2.1.0
• Added support for webdatasets (only recommended for large-scale distributed training)
• Improved per-question logging when training from scratch
• Added option to compile encoder for max speed (not recommended for finetuning, only for pretraining).
• Deprecates TensorFlow. The CS research community focuses on PyTorch and new frameworks like JAX.

Full Changelog: v1.0.5...v2.0

v1.0.5 - Finetuning Improvements, Max-ViT

What's Changed

Major improvements to finetuning.

• Now supports finetuning resnet and maxvit-tiny! This previously only worked as static pretrained models.
• Now specify num_blocks instead of num_layers, reflecting the network structure (effnet, resnet50, maxvit)
• Now you can optionally keep batchnorm layers always trainable

Small changes:
Update dependency to use galaxy-datasets==0.15.0
Replace multi-class example with new galaxy-mnist dataset
Fixes #108

The next release will update timm to 0.9ish
Pytorch 2.1.0 already works, and the next release will also reflect that

Full Changelog: v1.0.4...v1.0.5

Color and multi-class support

This minor release adds pretrained color models (i.e. with 3 channels) and also adds support for multi-class finetuning.

What's Changed

• sync by @mwalmsley in #92
• 1.0.3 version bump by @mwalmsley in #99
• Rebuild docs by @mwalmsley in #103
• Add color weights, trivial QoL changes by @mwalmsley in #104
• Add multiclass example by @mwalmsley in #105
• Add multiclass support by @mwalmsley in #106

Full Changelog: v1.0.3...v1.0.4

JOSS paper release

Minor release marking Zoobot's state on acceptance of the JOSS paper.

What's Changed

• Finetune v1 onto docs by @mwalmsley in #85
• Dev by @mwalmsley in #87
• sync by @mwalmsley in #93
• Delete requirements.txt by @mwalmsley in #96
• Bring paper fix from docs to main by @mwalmsley in #97
• Highlight Colab notebook by @mwalmsley in #98

Full Changelog: 1.0.2...v1.0.3

v1.0.2 - Lighning v2 and galaxy-datasets updates

• Lightning v2 now correctly uses early stopping
• galaxy-datasets version bumped to correctly ignore alpha channel of png images

What's Changed

• Hotfix - add early stopping back, update docs by @mwalmsley in #91

Full Changelog: v1.0.1...1.0.2

v1.0.1 - Lightning v2.0 support

Small incremental release adjusting the LightningModule hooks used in order to support Lightning v2. Lightning v2.0.0 (currently latest) is now the minimum required version.

What's Changed

• Update with Maja's changes by @mwalmsley in #89
• Support Lightning v2.0.0 by @mwalmsley in #90

Full Changelog: v1.0.0...v1.0.1

v1 release

This release completely rewrites the API and documentation based on user feedback during the one-year beta period.

Major changes include:

• New API for finetuning
• Shift to PyTorch/Lightning
• Support for all timm models
• Docs refocused on finetuning
• Pretrained model library added
• Data loading refactored to albumentations and mwalmsley/galaxy-datasets

Documentation is here. Installation and quickstart are on the README.

Thank you all for the help!

What's Changed

• Update refactor with latest changes by @mwalmsley in #15
• Dockerize zoobot for pytorch and tensorflow versions by @camallen in #14
• Refactor pytorch datasets by @mwalmsley in #16
• missed a rename, thanks cam by @mwalmsley in #17
• allow multiple catalog paths to be passed by @camallen in #19
• allow install of git based packages (pytorch_galaxy_datasets) by @camallen in #20
• Bring old dev branch up-to-date by @mwalmsley in #23
• Update finetuning_advanced example by @mwalmsley in #24
• PL logging by @mwalmsley in #22
• Improve data files for docker by @camallen in #21
• allow checkpointing setup to be customized by @camallen in #26
• Include Cam's latest features on generic dev by @mwalmsley in #27
• improve docker setup by @camallen in #28
• Trivial updates by @mwalmsley in #29
• Refactor PyTorch for explicit Lighnting/Wandb hyperparameters by @mwalmsley in #30
• Adding Zoobot to pypi by @mwalmsley in #31
• add gh action CI system by @camallen in #34
• add gh action to publish package to pypi by @camallen in #33
• remove travis CI integration by @camallen in #35
• Deprecate TFRecords by @mwalmsley in #32
• Fix doc build, tweak readme by @mwalmsley in #40
• Trivial typing change left behind by @mwalmsley in #55
• Create CODE_OF_CONDUCT.md by @camallen in #46
• Add PyTorch Finetuning Capability, Examples by @mwalmsley in #59
• Benchmarks by @mwalmsley in #58
• remove extra _non-star label for artifact task by @camallen in #60
• allow more params to finetuning via config object by @mwalmsley in #63
• add wandb logging, freeze batchnorm by default by @mwalmsley in #62
• Make sure TF and Torch versions have similar performance by @mwalmsley in #64
• Big messy PR of misc. improvements by @mwalmsley in #65
• use correct torchvision 0.13.1 package by @camallen in #69
• Update with previous commits by @mwalmsley in #72
• use latest galaxy-datasets package by @camallen in #75
• add timm package for pytoch cuda by @camallen in #77
• Docs by @mwalmsley in #78
• Merge pull request #78 from mwalmsley/docs by @mwalmsley in #79
• Finetune v1 by @mwalmsley in #80
• Joss by @mwalmsley in #81
• Syncing branches by @mwalmsley in #83
• Update finetune-v1 with docs changes by @mwalmsley in #84
• Finetune v1 onto dev by @mwalmsley in #86
• Add tensorboard writer callback to pytorch by @maja-jablonska in #54
• V1 release by @mwalmsley in #88

New Contributors

• @maja-jablonska made their first contribution in #54

Full Changelog: v0.0.3...v1.0.0

Docs pre-release

Improved documentation and refactored train API (pytorch).

Awaiting results from several segmentation experiments ahead of public release (inc pytorch version).

v0.0.2

Polish throughout and add new features. See #2 for a full description.

v0.0.1

Initial release.

This had enough documentation and code to replicate the DECaLS model and make predictions. There are a few minor missing arguments and similar typos that you might have stumbled into, because I made some last minute changes without updating the docs, but everything worked with a little stack tracing.