The author fixed a bug in calculating Acc.
HuiZeng opened this issue · comments
We recently found an implementation bug in calcualating the Acc indexes which were slightly different from the definition in our paper.
Fortunately, this bug does not affect any of the conclusions in our paper.
We have corrected this bug in the Matlab code and updated the corresponding values in the revised paper using blue-color font.
We apologize for the confusion to readers. The detailed changes in code can be found here:
5818679#diff-d2f9e0c4e433d155b38659652521d46a
The original code actually calculates the following Acc:
rather than the real definition in our paper:
Hi,I found that the acc2/5 is lager than the acc3/5 at the line VEN of the table 5 in your updated paper. Is that right? I think it is impossible according to definition of ACC in your paper.
Hi,I found that the acc2/5 is lager than the acc3/5 at the line VEN of the table 5 in your updated paper. Is that right? I think it is impossible according to definition of ACC in your paper.
Hi， it is possible.
Let me give one example, suppose we have only one test image and the returned 4 crops have ranks {1, 10, 2, 4} in groundtruth, then:
acc2/5 is (1+0)/2=0.5,
acc3/5 = (1+0+1)/3 = 0.67,
acc4/5=(1+0+1+1)/4=0.75.
Thanks for reply. I get that.