Table 3 PSNR comparison for different methods on each dataset, along with the time required for 10 training updates (in seconds) with a batch size of 2.
From: Deeply supervised two stage generative adversarial network for stain normalization
Method | TUPAC-2016 | MITOS-ATYPIA-14 | ICIAR-BACH-2018 | MICCAI-16-GlaS | Time (s) |
|---|---|---|---|---|---|
Macenko | 16.108±0.017 | 19.231±0.031 | 22.108±0.236 | 19.714±0.397 | – |
Reinhard | 15.088±0.051 | 17.401±0.111 | 21.752±0.068 | 18.650±0.252 | – |
Vahadane | 15.057±0.010 | 20.520±0.002 | 23.547±0.006 | 25.665±0.061 | – |
StainGAN | 25.732±0.444 | 24.419±0.519 | 26.493±0.528 | 24.196±0.521 | 2.208 |
SAASN | 36.399±1.178 | 29.053±1.874 | 33.484±2.201 | 26.899±1.117 | 6.300 |
CAGAN | 33.264±1.159 | 30.531±0.679 | 32.723±0.935 | 27.915±0.447 | 1.954 |
DSTGAN | 39.130±0.844 | 37.628±2.635 | 38.399±1.750 | 34.147±0.892 | 3.274 |
