saadwazir / HistoSeg

HistoSeg is an Encoder-Decoder DCNN which utilizes the novel Quick Attention Modules and Multi Loss function to generate segmentation masks from histopathological images with greater accuracy. This repo contains the code to Test and Train the HistoSeg
18 stars 4 forks source link

Dice's calculation is obviously wrong #4

Open Frank-Star-fn opened 1 year ago

Frank-Star-fn commented 1 year ago

In the GlaS dataset, your IoU is only 76.73. According to the mathematical definition of IoU and Dice coefficient, your Dice is absolutely impossible to reach 99.09, so your Dice coefficient calculation is obviously wrong. You should check your code implementation of Dice coefficient calculation.

在GlaS数据集里,你的IoU只有76.73,根据IoU和Dice系数的数学定义,你的Dice是绝对不可能达到99.09的,所以你的Dice系数计算显然是错误的。你应该检查一下你关于Dice系数计算这部分的代码实现。

Frank-Star-fn commented 1 year ago

formula: IoU=|AB|/(|A|+|B|-|AB|) Dice=2|AB|/(|A|+|B|)

Frank-Star-fn commented 1 year ago

with the formula: IoU=|AB|/(|A|+|B|-|AB|) Dice=2|AB|/(|A|+|B|)

we can get: Dice=2*IoU/(1+IoU)

your IoU is 76.73%, so your Dice should be about 86.833%

yese12 commented 1 month ago

Indeed, as of now, Dice still finds it difficult to reach the incredible number of 99