master
Migrate optimizers to torch (#1547)
Initializers to Torch (#1549)
Backported updates: - Docs: Centralize sphinx requirements - Logging: Handle Torch warnings and logging within our logging module - Logging: More robust array logging handling - Alignments: Store arrays as list in alignments file - Alignments: Minor constants + Pose updates - Add insightface_resnet network - Update requirements
Completely remove Keras from loss calculations (#1545)
Train: Migrate data loading to Torch (#1540)
bugfix: TFLambdaOp to ScalarOp on legacy update
bugfix: train data loading - resize function for arbitrary channels
logging - Add standardized class __init__ log function - Import logger from lib.__init__ to prevent custom log level errors when running non-fs scripts
Faceswap 3 (#1516)
Alignments: Migrate typed-dicts to dataclasses
Replace imageio with pyav/ffmpeg-binaries (#1534)
Extraction Overhaul (#1533)
bugfix: Loss function device allocation
convert bugfixes: - ffmpeg writer: Exit audio buffer when no audio available - convert: correctly apply y-offset - Fix color adjust when "none" selected