KeyPoint question

[wenhuach]

Hi, May I ask some questions related to keypoints? I am debugging function fastext_inner_loop_12 in FASTex.cpp. I have the following questions, 1 If the keypoint type is SEK, the variable "vmin" only considers part of Pd or Pb, why? 2 The keypoint score you used is the min(abs(Ix - Ip)), right? It's different with what you said in your paper. 3 May I know the reason why you check the corner and cornerout points? 4 If the current pixel value is 164, the circle 12 pixel values are 93, 99,74,205,151,188,188,168,20,19,20,100 respectively, and the thereshold is 12, then it will be considererd as a SBK. But it contains all the three types Pb,Ps and Pd, while the paper said it should be contain only 2 types, (Pb and Ps) or （Pd and Ps）. Thanks in advance. If I am wrong, please correct me.

[MichalBusta]

Hi, On 07/21/2016 11:00 AM, wenhuach wrote:

Hi, May I ask some questions related to keypoints? I am debugging funtion fastext_inner_loop_12 in FASTex.cpp. I have the following questions, 1 If the keypoint type is SEK, the variable "vmin" only considers part of Pd or Pb, why? 2 The keypoint scrore you used is the min(abs(Ix - Ip)), right? It's different with what you said in your paper.

2: yes, you are right - you found the bug in paper - there should be min instead of max.

• if you have the dark keypoint, you will the darkest pixel from the bright set, and the score is given by: Ix - min(Ip_min) - that is the reason why the "vmin" only considers part of Pd or Pb

3 May I konw the reason why you check the corner and cornerout points?

what is corner-out point? you mean bends?

Michal

Thanks in advance. If I am wrong, please correct me.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/MichalBusta/FASText/issues/9, or mute the thread https://github.com/notifications/unsubscribe-auth/AD6jsMV270hg5f2kyfxZP7ihSBYbosYmks5qXzVIgaJpZM4JRlZf.

[wenhuach]

1 on 1,If the current pixel value is 66, the circle 12 pixel values are 100,93,88,158,149,188,168,103,96,79,83,33,and the thereshold is 12. in this case, there are two problems. The first one is the code takes it as a SEK,but it cantains Pb and Pd. The second one is the vmin is 88, the code only considers the first 9 points, not update it afterwards, there is a 79. 3 on 3 ,I mean line 164 in FASTex.cpp, funtion getCrossCorner12. 4 on 4 , the code haven't checked every "or" operation in line 408-421 in FASTex.cpp, if d==3, then it's definitly not a text keypoint based on your paper. Also the code just use variable "same" in fucntion fastext_inner_loop_12 to count the change numbers , I guess that's why the problem happens. One simple solution maybe be can be like this, insert the following code before line 408 in FASTex.cpp int tmpd = 0; for (int i = 0; i < 12; ++i) { tmpd |= tab[ptr[pixel[i]]]; } if (tmpd == 3) { continue; }

[MichalBusta]

On 07/22/2016 03:45 AM, wenhuach wrote:

1 on 1,If the current pixel value is 66, the circle 12 pixel values are 100,93,88,158,149,188,168,103,96,79,83,33,and the thereshold is

12. There are two problems. The first one it the code take it as a SEK,but it cantains Pb and Pd. The second one is the vmin is 88, the code only consider the first 9 points, not update it afterwards, there is a 79.

can you send me image cut?

• from the example, the keypoint is SEK: threshold value is 78, there are all pixels brighter except 33 = SEK?

3 on 3 ,I mean line 164 in FASTex.cpp, funtion getCrossCorner12. 4 on 4 , the code haven't checked d = 3 in line 403-421 in FASTex.cpp, also the code just use variable "same" in fucntion fastext_inner_loop_12 to calculate the change numbers , that's why I guess the prblem happens.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/MichalBusta/FASText/issues/9#issuecomment-234433480, or mute the thread https://github.com/notifications/unsubscribe-auth/AD6jsK93ZN7dz7qOoLEPcNFnT78WRGYtks5qYCDBgaJpZM4JRlZf.