dannce v1.2.0

This is a major update introducing new features and codebase improvements.

Contributors: Diego Aldarondo, Timothy Dunn

New Features

• Support for setups that use mirrors to provide multiple views with just a single camera.
• Multi-animal COM detection
• Support for training over very large motion capture datasets via .npy volume caching
  • Subsets of training frames can be sampled by setting num_train_per_exp
• Support for video file chunks of variable length
• Multi-gpu training, set via the multi_gpu_train config parameter
• Slurm scripts for automation, parallelization, and hyperparameter grid searches.
• Train/validation splits can now be set reproducibly using the data_split_seed config parameter
• Ability to ignore specific landmarks during training via the drop_landmark config parameter
• Support for validation over a dedicated recording via the valid_exp config parameter
• Random view sampling augmentation (on by default)

Codebase Improvements

• Additional documentation and typing hints
• Faster loading of training frames
• Improvements to metadata saving
• Refactoring interface.py, inference.py, and generator.py
• Reduced the repository download size by removing large files from the git history.