search

cdcseacave / openMVS

open Multi-View Stereo reconstruction library
http://cdcseacave.github.io
GNU Affero General Public License v3.0
3.3k stars 905 forks source link

Is there any limitaions on input 3d structre which can be textured properly? #409

Closed sathishkumark27 closed 5 years ago

sathishkumark27 commented 5 years ago

Hi all, I am using openMVG and opemMVS to 3D texture on Ubuntu16.04. I have given some complex outdoor scene(images attached with this issue) for reconstruction and texturing using complete pipeline python script "MvgMvs_Pipeline.py" but could not able to get the dens point-cloud and texture correctly. Are there any limitations on the input scene for correct texturing and reconstruction?

Steps to reproduce the problem

Running the pipeline with below images will not reconstruct and texture correctly. I have used around 1200 images of the below scene but could not able t get the correct reconstruction

frame0 frame1 frame2.

Any help greatly appreciated.

cdcseacave commented 5 years ago

the only limitation is set by the amount of memory you have on your system, for 1200 images you probably need 128GB RAM

sathishkumark27 commented 5 years ago

I have ran the pipeline overnight with 16GB of RAM, 8 cores. Can you please tell why the memory matters? Even if we have less memory it should construct property but takes lot of time right?

On Tue, Feb 5, 2019, 12:43 PM cDc <notifications@github.com wrote:

the only limitation is set by the amount of memory you have on your system, for 1200 images you probably need 128GB RAM

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cdcseacave/openMVS/issues/409#issuecomment-460776464, or mute the thread https://github.com/notifications/unsubscribe-auth/Ad1Zb5X6N6Kp9fBX6FodFoiiIQsTaq0Nks5vKd76gaJpZM4ajxXg .

cdcseacave commented 5 years ago

Wrong, some steps require looking at all images, triangles, etc at the same time, so they need to fit in memory all at once. Try increasing the memory size, or split the scene in smaller subscenes and process them individually. OpenMVG supports that.