Using Colmap 2.1 dev with CUDA (*) on Windows 7 (**)
I have 603 pictures taken with a GoPro HERO4 of the inside of a theatre (jpgs, 4000*3000 pixels)
Automatic reconstruction seems to work OK,
I adjusted máximum image size to 4000 instead of the default 3200 and after succesfully extracting and matching features, I got a good looking sparse point cloud in two separate models.
I tried dense reconstruction, succesfully undistorted a set of 318 Images corresponding to the main model.
All the resulting undistorted images are 6395 x 3781 pixels
However when I get to Stereo stage, the program hangs after 2 minutes.
In the Log I see:
Reading configuration...
================================================
Processing View 1/319
================================================
Reading inputs...
It stays like that for a while, memory starts to fill and right before it closes:
PatchMatch::Problem
---------------------------------
ref_image_id: 238
src_image_ids: 157 160 (...) etc
PatchMatch::Options
================================================
gpu_index: 0
(...)
filter_geom_consistency_max_cost: 1
At this moment a message appears in the colmap.exe Console:
out of memory in C:/Users/jachoenh/Development/colmap/src/.\nvs\gpu_mat.h at line 151
And right after colmap closes, a last error for a fraction of a second before the Console closes too:
QObject::^QObject: Timers cannot be stopped from another thread
I tried the previous dev versión (Commit 308d5c1 on 2017 04 24 with CUDA) stable versión (COLMAP 2.1), in both cases RAM fills at a linear rate from 2 to almost 8 GB, stays flat, and then closes, taking about 8 minutes to hang.
Any ideas?
Best
(*) (Commit b2d00e1 on 2017-05-11 with CUDA)
(**) Windows 7 Ultimate SP 1 64 bits, Intel Core 2 Quad, 8 GB RAM, Geforce GTX 1060 6 GB, CUDA 8.0.0 driver (6.14.13.8205), Graphics Controller 382.05