Model Output Visualization for PyTorch Models by n-poulsen · Pull Request #2799 · DeepLabCut/DeepLabCut

This pull request implements methods to visualize maps output by PyTorch models (scoremaps, location refinement fields and part-affinity fields) as they are for TensorFlow models.

API Methods Implemented for the PyTorch Engine

  • extract_maps
  • visualize_scoremaps
  • visualize_locrefs
  • visualize_paf
  • extract_save_all_maps

UX Improvements

  • only start training pose models when epochs > 0
  • get_model_snapshots: when no snapshots were found, the snapshot manager's snapshot prefix was obtained incorrectly
  • fine-tuning SuperAnimal detectors: added key_metric so the best detector si saved