Flowmatching

Training and comparison of flowmatching w.r.t. DDIM diffusion.

Done on EchonetLVH dataset using zea implementations

• dataloader.py contains a simple tensorflow loader
• train.py trains flowmatching/diffusion model and stores to checkpoints folder.
• In eval_files folder, for after training:
  • plot_val_nv_loss.py plots training statistics of both models.
  • sample.py plots some samples vs num_steps for both models.
  • dps_comparison.py gives a comparison for DPS inpainting for a single evaluation.
  • dps_scanlines.ipynb loops over some samples and reports the DPS loss of both models.