16
General / Re: How to limit number of cpus used by Photoscan?
« on: June 18, 2018, 03:50:19 PM »
Maybe it can be a workaround for win 10, but I use Photoscan on Linux mainly.
Show Posts
This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.
17
General / Re: Example of markers
« on: June 18, 2018, 03:47:56 PM »
I have tried to use recommended markers(see attachment), but false positive rate is too high to use it, so for now seems this feature is useless.
18
Python and Java API / What is track_id?
« on: June 18, 2018, 12:48:03 PM »
In docs I can see only:
track_id Track index in PointCloud.Point and PointCloud.Projection
http://www.agisoft.com/pdf/photoscan_python_api_1_4_2.pdf
Is it just unique indentificator of point or is it related to some consept of 'track'?
Also what is meant by `Point valid flag`?
track_id Track index in PointCloud.Point and PointCloud.Projection
http://www.agisoft.com/pdf/photoscan_python_api_1_4_2.pdf
Is it just unique indentificator of point or is it related to some consept of 'track'?
Also what is meant by `Point valid flag`?
19
General / How to limit number of cpus used by Photoscan?
« on: June 18, 2018, 12:42:01 PM »
How to limit number of cpus used by Photoscan?
20
General / Re: Aligh photos accuracy and error measured by gcp
« on: June 01, 2018, 05:00:29 PM »
What do you mean by 'similar results'? I'm expecting to get worser results in terms of measured error by gcp points (that don't participate in optimization procedure) when I use low accuracy setting, is this right?
By 'marker projections' and 'tie point projections' setting you mean Marker Accuracy and Tie point accuracy in Image Coordinates Accuracy?
Photoscan manual have very little info about tie point accuracy:
Also I can't find description of how Camera Accuracy and Marker Accuracy can influence on error, as I understand when I set lower values I can get better results(tested on Highest settings), I have tested [1,0.1,0.01,0.001] with adaptime camera fitting that leads to [2.65m,0.2m,0.15m,0.149m] error, on 0.001 error stagnates, can it be harmfull to set so low values? And seems Camera accuracy deg don't affect results.
Also seems playing with Reference Setting and Optmize Camera give me less error than camera calibration procedure.
Quote
If you use Lower accuracy with the same Tie Point Accuracy value, it would mean that due to large size of the key points (and therefore matching points) the referencing and optimization would mostly rely on the coordinate information of the cameras and markers.In my case gcp markers are unchecked and don't used in optimization procedure as I understand. If I understand you correctly in case of low accuracy setting (with same tie point setting as on high setting) model less rely on matched points, but why this is good? One possible reason I can imagine is that on high accuracy model more rely on matched points, but matched points contain errors and that leeds to bigger error.
By 'marker projections' and 'tie point projections' setting you mean Marker Accuracy and Tie point accuracy in Image Coordinates Accuracy?
Photoscan manual have very little info about tie point accuracy:
Quote
PhotoScan matches images on different scales to improve robustness with blurred or difficult to match
images. The accuracy of tie point projections depends on the scale at which they were located.
PhotoScan uses information about scale to weight tie point reprojection errors. In the Reference pane settings dialog
tie point accuracy parameter now corresponds to normalized accuracy - i.e. accuracy of tie point detected
at the scale equal to 1. Tie points detected on other scales will have accuracy proportional to their scales.
This helps to obtain more accurate bundle adjustment results. On the processing parameters page of the
report (as well as in chunk information dialog) two reprojection errors are provided: the reprojection error
in the units of tie point scale (this is the quantity that is minimized during bundle adjustment), and the
reprojection error in pixels (for convenience). The mean key point size value is a mean tie point scale
averaged across all projections.
Also I can't find description of how Camera Accuracy and Marker Accuracy can influence on error, as I understand when I set lower values I can get better results(tested on Highest settings), I have tested [1,0.1,0.01,0.001] with adaptime camera fitting that leads to [2.65m,0.2m,0.15m,0.149m] error, on 0.001 error stagnates, can it be harmfull to set so low values? And seems Camera accuracy deg don't affect results.
Also seems playing with Reference Setting and Optmize Camera give me less error than camera calibration procedure.
21
General / Re: Aligh photos accuracy and error measured by gcp
« on: May 31, 2018, 10:54:27 AM »Quote
The error of your check points (= GCPs that have not been used as control points) are influenced by quite a number of factors, hence it will not simply linearly change depending on chosen quality settings in the photo alignment step.
What factors can cause such non-linear dependence? At first I was thinking about that too many distortion coefficients can cause 'overfitting' and it will make error higher when I run experiments with adaptive flag on, but then I runned experiments with fixed number of distortion parameters result were about the same and even on lowest setting error were even smaller .
22
General / Reproducibility of results
« on: May 30, 2018, 09:22:42 PM »
Can I get fully reproducible results in Photoscan? As I can see some procedure (maybe optimization) have some randomness, so can I fix random seed somehow to get fully reproducible report?
23
General / Aligh photos accuracy and error measured by gcp
« on: May 30, 2018, 03:06:12 PM »
As I can see in Alignment parameters section in Photoscan manual
http://www.agisoft.com/pdf/photoscan-pro_1_4_en.pdf
As I understand accuracy setting can be interpreted as image downsize factor:
Highest x2 up
High original size (?)
Medium x2 down
Low x4 down
Lowest x8 down (?)
Is it right?
Also when I measure accuracy of model by gcp points (that was not included in alighnment process) I expect that higher accuracy will give smaller error, but seems dependense is not linear and at lowest setting I get smallest error and that is weird, can someone confirm that in some cases it can be the case? Also I have tried to run this experiments without Adaptive camera model fitting and this strange behaviour preserves.
http://www.agisoft.com/pdf/photoscan-pro_1_4_en.pdf
Quote
Accuracy
Accuracy
Higher accuracy settings help to obtain more accurate camera position estimates. Lower accuracy
settings can be used to get the rough camera positions in a shorter period of time.
While at High accuracy setting the software works with the photos of the original size, Medium setting
causes image downscaling by factor of 4 (2 times by each side), at Low accuracy source files are
downscaled by factor of 16, and Lowest value means further downscaling by 4 times more. Highest
accuracy setting upscales the image by factor of 4. Since tie point positions are estimated on the
basis of feature spots found on the source images, it may be meaningful to upscale a source photo
to accurately localize a tie point. However, Highest accuracy setting is recommended only for very
sharp image data and mostly for research purposes due to the corresponding processing being quite
time consuming.
As I understand accuracy setting can be interpreted as image downsize factor:
Highest x2 up
High original size (?)
Medium x2 down
Low x4 down
Lowest x8 down (?)
Is it right?
Also when I measure accuracy of model by gcp points (that was not included in alighnment process) I expect that higher accuracy will give smaller error, but seems dependense is not linear and at lowest setting I get smallest error and that is weird, can someone confirm that in some cases it can be the case? Also I have tried to run this experiments without Adaptive camera model fitting and this strange behaviour preserves.
24
General / Re: Camera calibration for UAV
« on: May 30, 2018, 12:34:51 PM »
Also if I have 2 cameras of same model is it reasonable to use different calibration for each of them or it's ok to use averaged calibration?
25
General / Camera calibration for UAV
« on: May 29, 2018, 12:24:26 PM »
I have a camera that used in UAV flights and I want to calibrate it.
In flight camera focus is set towards infinity, so when I try to calibrate camera I use this setting, but in this setting chessboard looks blured when I move camera in front of screen to cover full screen frame, calibration process seems works fine but do I do everything right?
Also can you clarify in Camera Calibration settings what is difference between Focal length (mm) and f parameter (according to documentation it's in pixel and also called focal length), is the same focal length just in different units or it's mean something different?
In flight camera focus is set towards infinity, so when I try to calibrate camera I use this setting, but in this setting chessboard looks blured when I move camera in front of screen to cover full screen frame, calibration process seems works fine but do I do everything right?
Also can you clarify in Camera Calibration settings what is difference between Focal length (mm) and f parameter (according to documentation it's in pixel and also called focal length), is the same focal length just in different units or it's mean something different?
26
General / Re: Measuring error by gcp points
« on: May 28, 2018, 08:29:46 PM »
Solved. It was because 2 gcp points were not georeferenced on images.
27
General / Measuring error by gcp points
« on: May 28, 2018, 07:07:41 PM »
I'm trying to measure error by gcp points, there are 3 of them, but for some reason error statistics is empty for 2 of them. Is it mean that error is zero? or is it indicates some problem?
28
General / Re: Is it possible to import / export only markers?
« on: May 28, 2018, 06:40:25 PM »
As I can see when I export markers .xml contain also additional information like camera parameters, can you confirm that it will not overload current project parameters when I import saved markers?
29
General / Re: Estimated altitude less then ground level
« on: May 28, 2018, 05:59:35 PM »
The problem was that X and Y coordinates were swapped.
30
General / Re: Estimated altitude less then ground level
« on: May 28, 2018, 02:56:00 PM »
In my case it's not DJI drone and as I can see by clicking export exif there is no meta info in images, so I load positions of cameras via import on refrence tab.
When I perform step Aligh Photos error between source and estimated camera positions in order of meters, but when I load gcp point and then set x,y coordinates of it on image, I get high error on Z coordinate (source altitude 150m and estimated 382m)
When I perform step Aligh Photos error between source and estimated camera positions in order of meters, but when I load gcp point and then set x,y coordinates of it on image, I get high error on Z coordinate (source altitude 150m and estimated 382m)
