• Segmentation and classification accuracy metrics
  • Precision
  • Recall
  • F1-Score
  • Dice
  • IOU
  • Pixel Accuracy
  • Aggregated Jaccard Index (AJI)
• Realistic image synthesis metrics
  • MSE
  • SSIM
  • Inception Score
  • FID Score
  • SWD

Please use the conda environment file.

conda env create -f environment.yml
conda activate statistics_env

To compute the segmentation/classification scores between two datasets, where images of each dataset are contained in an individual folder:

python ComputeStatisctics.py --gt_path path/to/ground-truth-masks 
                             --model_path path/to/masks-generated-by-model
                             --output_path path/to/output

Arguments:

• Required
  • gt_path: path to ground-truth masks
  • model_path: path to masks generated by the model
  • output_path: path to save output csv files
  • mode: mode of the statistics computation including Segmentation (for computing segmentation accuracy metrics), ImageSynthesis (for computing realistic image synthesis metrics), All (for computing segmentation and image synthesis metrics)
  • raw_segmentation: use this argument if the segmentation mask needs refinement
  • image_types: types of images