yfcc evaluation using sift

[panxkun]

Hi, appreciate your work very much!
I have trouble again reproducing the result in your paper. In YFCC eval, your code only get the acc result using flow, but now I want to use the sift method to test my code to ensure it is right, my code was mainly modified on your code. I get the result in this paper:

the main code as follow:

the function in this code are all copy from your script. maybe I wrongly use some parameters like ransac threshold or matcher, could you please tell me some detail about this evalution?

[fdarmon]

Hi,

If I remember correctly, the results on the paper were obtained with 2000 SIFT features. We used ratio test with ratio 0.8 and also bidirectional check.

François

[panxkun]

thanks, it works!