search

seravee08 / WideBaselineFeatureMatcher_PAMI

This repository contains .exe demo
GNU General Public License v3.0
37 stars 24 forks source link

“-noaf ” cannot work #7

Open brettzhao opened 5 years ago

brettzhao commented 5 years ago

As the "README" says, if I add "-noaf " after "-sift", l can get the result detected by traditional sift, rather than A-SIFT. But even if I add this parameter, the program still detects a-sift keypoints. Is there anything I forget to pay attention to or do I misunderstanding the meaning of the parameter? By the way, what are ".aff" files used for ?

seravee08 commented 5 years ago

As the page said, you may use C:/CODE -sift -noaf -indb C:/image_list.txt to disable the affine functionality. If you did it in this way, how did you determine that your results were still with affine? You might also want to play with the sift parameter file siftParam_settings.txt, which might be the real reason behind the abundant sift keypoints.

Affine sift features need to save out affine information besides the standard sift so it can be read back for further processing. If I put both sift and affine inside a single file, it is no longer compatible with main-stream software like VisualSfM. So I split it into two files.

brettzhao commented 5 years ago

thank you for your answer,I found out the reason. But I cant open changchangwus pdf when I want to change the sift parameters, would you please explain the meaning of the number in siftParam_settings.txt

brettzhao commented 5 years ago

By the way，how can this exe save result images automaticly?

seravee08 commented 5 years ago

It is recommended that you rear the original sift and affine-sift paper. As a quick guide, tilt controls the affine simulations, fo is short for first octave. t is the parameter that affects the results most, a reasonable number would be something between 0.2 - 0.002