v0.14.2

• new backbones available (ConvNeXt, ConvNeXtV2, HRFormer)
• new checkpoints available (clipconvnextbase, convnextv2base, hrformerbasecat, shufflenetv2k30*, swin_l_input_upsample)

v0.14.1

• flexible compatibility with recent PyTorch versions
• option to resume optimizer's state from checkpoint
• AdamW optimizer, linear learning rate scheduler
• fast scaling as default for data augmentation
• improve configuration and display for eval

v0.14.0

• create openpifpaf-vita package, with separate development from openpifpaf
• deploy openpifpaf-vita guide