dice_average.m and compute size of polyp about caranet

[ShuyueG]

Hi! How did you use the function, and what errors have you got? Please carefully read the annotations in the code. Note: the arrays AreaR and Dice should have the same length. For i-th case, AreaR(i) is the area ratio (see code comments for its computation) of objects in the image of i-th segmented ground truth (the binary label), Dice(i) is the Dice coefficient computed by using the i-th predicted segmentation result and its ground truth. bin_width depends on the distribution of AreaR.

To your second question, yes, the x-axis means the relative size = # of pixels of polyp / # of total pixels of the image.

[hybdxxw]

Thanks for the reply.
I add these code into main.m , there are no errors, but the generated pic is just one straight line
[AreaR]=deal(zeros(1,imgNUM));
[r,c] = size(resmap);
AreaR(i) = sum(resmap(:))/(r*c);
dice_average(AreaR,threshold_Dic,0.05)

I also input just AreaR,threshold_Dic in the function for a test ,but the results is the same.

[AngeLouCN]

Hi, you don't need to add those code into main.m, because they are for different purposes. The matlab codes in fold eval are used to evaluate the segmentation results by 6 different measurement metrics. The dice_average are used to analyze the relationship between object size and dice accuracy.

[hybdxxw]

Thanks for the reply.
But how can I input data and run dice_average ? For example, if I want to analyze the relationship between object size and dice accuracy in CVC-ClinicDB, does it means I need to get the dice for every picture in test dataset ,area ratio of objects in images one by one and Width of bins for taking average(bin_width depends on the distribution of AreaR, what does this means?Do I need compute or just input 0.05) ,and put them in arrays, then run dice_average in command window?
Thank you very much for giving me your valuable time for these questions!

[ShuyueG]

Thanks for the reply. But how can I input data and run dice_average ? For example, if I want to analyze the relationship between object size and dice accuracy in CVC-ClinicDB, does it means I need to get the dice for every picture in test dataset ,area ratio of objects in images one by one and Width of bins for taking average --Yes, you're right.

(bin_width depends on the distribution of AreaR, what does this means?Do I need compute or just input 0.05)
-- no need to compute, just try 0.05 or other values. It depends on the range or density of the AreaRatio. The bin_width plays a role like the bin width when you draw a Histogram of the AreaRatio. It groups similar area ratios together and calculates their means.

,and put them in arrays, then run dice_average in command window? -- Yes, or put it in your codes; just like other Matlab functions.

[MELSunny]

dice average (mDice), please refer to the code from segmentation.
it sums the area of intersect and union, prediction (input), label(target) in all test images.
and then calculate (2 * total_intersection ) / (total_input + total_target)

https://github.com/open-mmlab/mmsegmentation/blob/master/mmseg/core/evaluation/metrics.py

#1