mapillary
Adds Dense SIFT feature type. http://www.vlfeat.org/overview/dsift.html
Hi, vl/dsift.h is already present in opensfm/src/third_party/vlfeat/vl/dsift.h They are not part of this pull req.
-sashi.
On Fri, Apr 20, 2018 at 8:39 PM, Russ [email protected] wrote:
@russkel commented on this pull request.
In opensfm/src/dsift.cc https://github.com/mapillary/OpenSfM/pull/236#discussion_r183199051:
@@ -0,0 +1,52 @@ +#include "types.h" +#include "dsift.h"
+#include
+#include + +extern "C" {
- #include "vl/dsift.h"
Are these files included in the pull req?
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/mapillary/OpenSfM/pull/236#pullrequestreview-114161644, or mute the thread https://github.com/notifications/unsubscribe-auth/Af6yBm19FiMlzHaRQAvhcVcqlfAf-KCaks5tqqnqgaJpZM4QVw9n .
Is there a chance you could resolve the conflict in the features.py file @sashichandra ?
@paulinus are you willing to merge this? Otherwise I'll pull this into my own working copy.
@russkel, I've resolved the conflicts on this branch and put the result in this branch https://github.com/paulinus/OpenSfM/tree/sashichandra-dsift
About merging it to master, the main concern is that Dense SIFT is not a suitable feature for SfM in general. The reason for that is that it extracts features on a fixed grid that is not adapted to the actual image texture. As a result, corresponding features in different images are not centered around the same real-world point. This adds substantial error to triangulation and resection algorithms to the point where they no longer work.
Testing this branch on berlin and lund dataset, I was unable to get good reconstructions. Only berlin yield a reasonable reconstruction after setting dsift_step to a small number so that the discretization error was reduced. That produced a lot more features which were then slow to match.
If there is a trick to make things work better, please let me know.
Hi @sashichandra!
Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file.
In order for us to review and merge your code, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.
If you have received this in error or have any questions, please contact us at [email protected]. Thanks!
