v0.11.1 - PyTTI LIVES AGAIN!

inplace operations were causing issues following upstream updates to pytorch.

v0.11.0 - Feature Parity!

This release includes fixes for the only remaining places where pytti-tools and pytti 5 beta still differed in behavior. Namely:

• show_palette - Previously would throw errors, workaround was to keep this disabled. Palette now displays correctly without errors.
• Limited Palette::init_image - Limited Palette wasn't properly loading init_image. Fixed! details below.

Details on the Limited Palette::init_image fix

Observed behavior was that Limited Palette generations with an init image would initialize to a black+white image, then the init_image would populate after a few frames generated. Cause was a bug that was introduced with the addition of gradient accumulation. Fixed by restructuring conditional statement in training code that was causing full bypass of update procedure if no Embedder object (i.e. CLIP model) passed to trainer. Etiology of this bug suggests there was probably also some difference in how motion transformations impacted animations for Limited Palette, and also accounts for difference in speed reported by users (since pytti-core was skipping the 201 update steps hard-coded into the PixelImage.encode_image(smart_encode=True) behavior).

v0.10.3 - Bug Fixes

• Fixed minor regression from 0.10.2
• Fixed show_palette bug that I could swear I've fixed three times already

Device Selection

• added support for device configuration (resolving forced usage of cuda:0)
• refactored image_model's
• added tests, docstrings, typehints
• isolated cutouts code to facilitate increased cutout control, future support for Dango's method, etc.
• POC'd backwards-compatible approach for adding config options using open_dict context manager
• fixed tests that used local paths

v0.10.1 - MMC perceptors added

Added support for loading perceptors via https://github.com/dmarx/Multi-Modal-Comparators/ . List of supported perceptors extended by about 30 or so and will continue growing indefinitely as support for more perceptors is added to mmc

v0.10.0 - Added Audio-reactivity

What's Changed

• Feat/audio input (rebased) by @dmarx in #140
• batbot URLs for coco artifacts fixes vqgan by @dmarx in #143
• added ViT-L/14_336px support by @dmarx in #148
• AV improvements by @dmarx in #150
• minor fixes by @dmarx in #151

Full Changelog: v0.9.1...v0.10.0

v0.9.1

What's Changed

• fixed logging bug in new update_func module by @dmarx in #137

v0.9.0

What's Changed

• added image_model validation to workhorse by @dmarx in #105
• More config validation by @dmarx in #106
• set restore from params by @usergenic in #112
• refactor loss configuration setup phase by @dmarx in #110
• VQGANImage refactoring by @dmarx in #111
• Fix video source animation by @dmarx in #130
• merge test for release by @dmarx in #135

New Contributors

• @usergenic made their first contribution in #112

Full Changelog: v0.8.1...v0.9.0

v0.8.1

What's Changed

• Structured configs2 by @dmarx in #90
• use pd.concat instead of append (perf and warning) by @ltearno in #91
• added use_tensorboard config option by @dmarx in #100

New Contributors

• @ltearno made their first contribution in #91

Full Changelog: v0.8.0...v0.8.1

v0.8.0

• Fixed a bunch of bugs from the refactor that went under the radar of the tests
• Fixed a bug that was causing PyTTI to not save the last frame it generated
• Added a bunch of CLIP models + automated support for anything the PyPI distributed clip package supports