cdcseacave
commented
4 years ago Usually this error is given when an image does not have any appropriate neighbors that allows for estimating a good depth-map (ex good baseline, good scale, angle, etc). In this particular case the problem seems to be very small baseline. Of course you can run the densification algo on these images as well, even if small baseline, but the result is in general very noisy due to high imprecision for far away points. Feel free to play with disabling the baseline check in the code (SelectViews()). And in general is good to use a small focal length, do not zoom but get closer to the scene.
On the other hand, the densification algorithm implemented in OpenMVS is for general images, while these seem to be stereo images. For stereo images I recently implemented a SGM algorithm inside OpenMVS, and you should give it a try.
sangukbo
CeciliaPYY
Hi,
I am trying to use DensifyPoinitCloud function, but I keep getting the following error.
The task that I am trying to do is the following.
I have done SfM using openMVG using 200 images. There were about 200000 landmarks generated in this process.
Among 200 images, I am interested in doing MVS with only two of the images ("left.png" and "right.png").
I modified sfm_data.json file from the openMVG such that it only contains two views that I am interested ("left.png" and "right.png"). I also included only two extrinsics for the views. I kept the "structure" and "control_points" in sfm_data.json the same (containing about 200000 landmarks as they were).
I converted the new sfm_data.json file to .mvs file. The process captures 4700 landmarks that are relevant to the two views.
Then I performed DensifyPointCloud. I got the error that says that image 0 and image 1 ("left.png" and "right.png") has no good images in view.
I am confused because I thought 4700 landmarks would be sufficient for performing DensifyPointCloud for two images. I performed DensifyPointCloud with about 1000 landmarks with two images before and it worked okay.
I used ./viewer to check if the camera direction was correct and it captured enough landmarks (or points). It seems like the camera view is okay. I have attached the view image as view.png.
I have included the two images and .mvs file just in case.
I am trying to use 2 images among 200 images (that I have used when I did sfm), because I only need small part of the 3-D model really accurate, and I do not need to do that for the whole set of 200 images. The accurate MVS computation for the small part of the 3-D model needs to be done fast enough.
I thought using 4700 landmarks would be enough to perform MVS with two images, because I have ran openmvs with 1000 landmarks using two images.
Could you please help me in finding out how to resolve this issue?
In addition, I was also wondering if there is a way to open and modify .mvs file. For instance, sfm_data.bin file created in openmvg can be converted into .json file and I could modify the file. I was wondering if this would be possible for .mvs file and if so, I would like to know how.
I have attached the view.png, left.png, right.png, scene.mvs, and log file in openmvs.tar.gz file.
openmvs.tar.gz
I am using the most recent version of openmvs on ubuntu 16.04 machine.
Thank you very much for your help.