v2.2.4

• major README overhaul
• enabled thingsvision.__version__ for automatic versioning

v2.2.3

• THINGSvision changed to thingsvision. This is more intuitive.
• We've added a low-memory option for feature extraction.
• extractor.extract_features() now takes extra optional arguments output_dir and steps.
• if output_dir is defined, features will be written to disk every steps mini-batches (default: 100).
• iteration across mini-batches got abstracted into the base extractor class.

v2.2.2

new patch version

v2.2.0

• modularized feature extraction (added more abstractions)
• added base extractor class from which all source-specific extractor classes are supposed to inherit
• mixin classes for both PyTorch and TensorFlow
• source-specific extractor classes that inherit from both the base extractor class and mixin classes like so

@dataclass(repr=True)
class TimmExtractor(BaseExtractor, PyTorchMixin):
    ...

• modified unittests according to most recent changes (not much has change here)

v2.1.4

updated patch version

v2.0.11

Update tests.yml

v2.0.9

• More abstractions (away from the extractor class)
• New custom models trained on EcoSet (https://www.pnas.org/doi/abs/10.1073/pnas.2011417118)
• A whole bunch of new pretrained CLIP models (https://github.com/mlfoundations/open_clip)

v2.0.7

• HDF5Dataset class to extract features for datasets with hdf5 files such as the NSD dataset (http://naturalscenesdataset.org/)

v2.0.6

Compatibility with most recent torchvision (0.13.) and timm (0.6.) versions fully enabled

What's Changed

• fix issue #65 by @LukasMut in #68

Full Changelog: v2.0.2...v2.0.6

v2.0.5

What's Changed

• Create LICENSE by @LukasMut in #1
• Reviewing test procedure by @matthew-brett in #5
• added Tensorflow support by @hahahannes in #10
• refactored tests & added directory structure; fixes #12 by @hahahannes in #13
• More vision tests by @hahahannes in #15
• added ecoset; fix #16 by @hahahannes in #17
• update tensorflow to 2.5.1 by @hahahannes in #19
• added test with batch sizes; fixes #20 by @hahahannes in #21
• added center cropping flag; fixes #25 by @hahahannes in #26
• add center cropping to readme; fix #27 by @hahahannes in #28
• User models by @hahahannes in #30
• added colab notebook; fix #31 by @hahahannes in #32
• updated notebooks by @hahahannes in #33
• updated vgg ecoset name; fix #34 by @hahahannes in #35
• updated notebook by @hahahannes in #39
• updated readme; fix #38 by @hahahannes in #40
• updated alexnet ecoset model weight path by @hahahannes in #43
• added more tf vs. torch tests by @hahahannes in #47
• transform pytorch tensor to numpy array; fix #48 by @hahahannes in #50
• added timm support; fix #54 by @hahahannes in #55
• aligned with requirements.txt by @PhilippKaniuth in #58
• Added missing equality sign by @PhilippKaniuth in #59
• added model source parameter; fixes #56 by @hahahannes in #57
• Remove circular imports error. by @florianmahner in #61
• Oop refactor by @LukasMut in #63
• added tqdm by @LukasMut in #64

New Contributors

• @matthew-brett made their first contribution in #5
• @hahahannes made their first contribution in #10
• @PhilippKaniuth made their first contribution in #58
• @florianmahner made their first contribution in #61

Full Changelog: https://github.com/ViCCo-Group/THINGSvision/commits/v2.0.2