/sn-calibration-segmentation

Primary LanguagePython

EVS Camera Calibration Challenge

This is a fork of the EVS Camera Calibration Challenge repository.

Usage

Train a new model using the src/segmentation directory.

Extract extremities keypoints using custom_extremities.py.

Overview of useful arguments:

  • -p: Output directory name for predictions (and masks).
    • Note: The final path for outputs will be extended: DIRNAME_np{args.pp_num_points}_r{args.pp_radius}_md{args.pp_maxdists} / args.split
  • --split: Select data split from base directory
  • --checkpoint: Model checkpoint for self-trained network.
  • --pp_radius: Post processing: Radius of circles that cover each segment. (Default changed from 6 to 4)
  • --pp_maxdists: Post processing: Maximum distance of circles that are allowed within one segment. (Default changed from 40 to 30)
  • --pp_num_points: Post processing: Number of keypoints that represent a segment.

Example:

sn-calibration/src> python custom_extremities.py -p pitchloc_train --checkpoint ../resources/train_59.pt --split test --pp_num_points 4