search

yosungho / LineTR

Line as a Visual Sentence: Context-aware Line Descriptor for Visual Localization (Line Transformer)
Other
238 stars 35 forks source link

Question about line homography estimation #5

Closed ChrisLLLLL closed 1 year ago

ChrisLLLLL commented 2 years ago

Thank you for your work. When i view your code,I noticed that there is no code for homography. Cloud you teach me how to achieve the result like Figure 5 in your paper？I have sent you an email. Looking forward to your reply.

yosungho commented 2 years ago

Hello, Chris. I used ground-truth line correspondences generated by a synthetic for the homography matrix. I also used 3D line projection to create ground-truth pairs for the ScanNet datasets. Furthermore, the color of the line pairs can be set using the 3D-lines' reprojection error given 6DoF poses for Oxford RobotCar dataset.

ChrisLLLLL commented 2 years ago

Thank you, but I want to know how to get the "correct matches (green) and the wrong matches (red) " in the matching of a pair of images. I could not find the code in LineTR.

yosungho commented 2 years ago

Hi Chris, sorry for the late reply. The correct and incorrect matches can be known with the homographic/geometric relationship (the first and the rest datasets in my paper). For example, when the lines on one image should be projected onto the lines on another image if their matches are correct. If their matches are incorrect, the reprojection error will be large. For the projections, you can use a known synthetic homography matrix or a perspective projection in 3D line maps. I recommend you to refer to the paper for more details. The code only contains the inference code.