search

openphotogrammetry / meshroomcl

MeshroomCL: An OpenCL implementation of photogrammetry with the Meshroom interface
Other
163 stars 8 forks source link

Multiview SterioCL has 101 problems #43

Closed Dobby-progam closed 2 years ago

Dobby-progam commented 2 years ago 
Writing photometric output for 960640750.JPG

==============================================================================
Processing view 101 / 101
==============================================================================

Reading inputs...

PatchMatch::Problem
-------------------
ref_image_idx: 100
src_image_idxs: 36 20 17 6 88 8 35 40 18 62

PatchMatchOptions
-----------------
max_image_size: 3200
gpu_index: 0
depth_min: 2.02196
depth_max: 6.46737
window_radius: 5
window_step: 1
sigma_spatial: 5
sigma_color: 0.2
num_samples: 15
ncc_sigma: 0.6
min_triangulation_angle: 1
incident_angle_sigma: 0.9
num_iterations: 5
geom_consistency: 0
geom_consistency_regularizer: 0.3
geom_consistency_max_cost: 3
filter: 1
filter_min_ncc: 0.1
filter_min_triangulation_angle: 3
filter_min_num_consistent: 2
filter_geom_consistency_max_cost: 1
write_consistency_graph: 0

PatchMatch::Run
---------------
OpenCL platform : AMD Accelerated Parallel Processing 
OpenCL device   : gfx902 
Initialization: 2.99105s
 Sweep 1: 15.367s
 Sweep 2: 17.1035s
 Sweep 3: 15.8111s
 Sweep 4: 16.8561s
Iteration 1: 65.1384s
 Sweep 1: 16.1526s
 Sweep 2: 16.9723s
 Sweep 3: 15.8775s
 Sweep 4: 17.3847s
Iteration 2: 66.3875s
 Sweep 1: 15.8669s
 Sweep 2: 17.1401s
 Sweep 3: 16.2977s
 Sweep 4: 17.1907s
Iteration 3: 66.496s
 Sweep 1: 14.2633s
 Sweep 2: 15.2819s
 Sweep 3: 15.6453s
 Sweep 4: 16.5248s
Iteration 4: 61.7157s
 Sweep 1: 15.8349s
 Sweep 2: 16.6432s
 Sweep 3: 15.0336s
 Sweep 4: 14.9508s
Iteration 5: 62.4629s
Filtering: 0.142245s
Total: 325.334s

Writing photometric output for 992787817.JPG
Elapsed time: 26.661 [minutes]
Reading workspace...
Reading configuration...
Configuration has 101 problems...

==============================================================================
Processing view 1 / 101
==============================================================================

Reading inputs...

as you can see the Multiview starts over I wonder if this is normal or not

I didn't change any settings

Dobby

revisionarian commented 2 years ago

Hi @Dobby-progam, yes this is normal behavior for the MultiviewStereoCL node. The phrase "101 problems" just means that there are 101 images to be processed, not that there is anything wrong. And the MVS processing is a two-pass algorithm, so you will see it count from 1 to 101 twice before it is done. You will notice that the "geom_consistency" field is set to "0" on the first pass, and "1" on the second pass. The second pass with geometric consistency enabled creates a more accurate reconstruction that is consistent between the depth maps from each view.

Dobby-progam commented 2 years ago

So it has to run 101 times and proses in that time 101 images Okay tanks for your response!

revisionarian commented 2 years ago

In your case, there are 101 images to be processed by the MultiViewStereoCL node. Each of the those 101 images will be processed twice by the MVS algorithm that creates depth maps. So, you will see a message like this:

==============================================================================
Processing view xx / 101
==============================================================================

a total of 202 times, and the total time for this step will be about one hour on your computer.