This is a repo. for evaluation metrics using Pytorch. The metrics.py is designed for evaluation tasks using two pytorch tensors as input. All implemented metric is compatible with any batch_size and devices(CPU or GPU).

y_pred << 4D tensor in [batch_size, channels, img_rows, img_cols]
y_true << 4D tensor in [batch_size, channels, img_rows, img_cols]

metric = MSE()
acc = metric(y_pred, y_true).item()
print("{} ==> {}".format(repr(metric), acc))

• python3
• pytorch >= 1.
• torchvision >= 0.2.0

• Image similarity

  • AE (Average Angular Error)
  • MSE (Mean Square Error)
  • PSNR (Peak Signal-to-Noise Ratio)
  • SSIM (Structural Similarity)
  • LPIPS (Learned Perceptual Image Patch Similarity)

• Accuray

  • OA(Overall Accuracy)
  • Precision
  • Recall
  • F1-score
  • Kapp coefficiency
  • Jaccard Index

• FID(Fréchet Inception Distance)

Our implementations are largely inspired by many open-sources codes, repos, as well as papers. Many thanks to the authors.

• Richard Zhang, LPIPS(https://github.com/richzhang/PerceptualSimilarity)
• Jorge Pessoa, SSIM(https://github.com/jorge-pessoa/pytorch-msssim)

This implementation is licensed under the MIT License.