3D photogrammetry guide

Please only post topics here related to metal detecting tips and tricks & D.I.Y
Post Reply
User avatar
Posts: 656
Joined: Thu Nov 23, 2017 7:00 pm
Has thanked: 1095 times
Been thanked: 405 times

3D photogrammetry guide

Post by Twit » Tue Mar 13, 2018 5:45 pm

Short intro to 3D photography.

Have to thank fisher 1266 for all of this for making aware of free software and its capabilities :D .

The download link is


I'm running it on a 2ghz (peak I think) 2gb ram tablet with 10gb disk space and it works on those minimum specs.

There are a series of tutorials and longer explanations online, and I'll list a few simpler links at the end for some depth, but here I am going to keep it very simple - the software is near automatic IF you start with the right input, which is good photos that the software can recognise and work with.

Here is the start of the discussion with fisher 1266 , and he provides a very good image of how the turntable and markers benefit from being set up.

viewtopic.php?f=42&t=98885&p=878626&hil ... ed#p878626

There are a few requisits for a good photo set:

The more in focus the better, phone cameras have fixed aperture, but digital cameras can benefit from decreasing it (larger fstop) . Phone cameras in general have no fixed or manual focus ( rooted it is possible I think), so each shot is autofocus hit and miss as you go along...keep trying for each shot until it seems ok.

The clarity is helped by tripod, delayed shutter release, good lighting, low iso etc.

Lighting. The idea is to have consistent lighting and avoid shadows or large changes in surface appearance that confuse the processing. A phone flash, which is sat next to the lense, leaves almost no visible shadows in a photo, but it does leave high reflections and over highlights certain areas. I managed to get some sets accepted using a phone fill flash, but only had two whole sets ( from different heights) processed correctly using a light tent, pictured below :


This was a few hours project, all thin ply, the turntable is a chair wheel swivel, the bracket at left holds the phone at four different angles, the light is a battery bar led one, the light tent is paper on wood frame, the top sheet is much finer paper to let in light. Turntable surface and background can be white or black, I am still working out which is best, as well as how to get a complete 360 image without the base. Total cost @ £10.

So you will be taking at least one circle of at least ten pictures ( though it is possible to do just say a front surface not 360) , better results are two rows of twenty or so each, and be asking the program to figure them into order to give a geometric 3D calculation of the position of the object in space. Sometimes it manages only part of a set, gives half a view etc. .

This first step is called the sparse cloud, which is made up of points shared by different photos that are recognised by the programme. It is the one you have to get right, as once the programme has properly placed all (or most) of the pictures, the rest is just refinement.

After the sparse cloud is judged ok by the user, you ask it to make a dense point cloud... which is as name suggests, reconstruction of many more common points.

The next step is to ask it to create a triangle mesh representing the surface between points.

Finally you ask it to colour in the mesh - textured mesh.

The only step that is difficult is the first, the rest work afterwards.

The download and installation of the program was fine ( @ 300 mb) , and straight off to try a set of photos.

Started freehand without turntable at night with whatever lamps and torches I could find, but it only accepted a few photos from each set ( e.g. three of twenty) ... so tried in daylight and a bit better, then with backdrops, masking, simple marking etc.... and still at best only half a model.... then put together a turntable and tried fisher's style of markings, bought a small tripod, and was getting near a single set accepted...sometimes... so decided the tripod was awkward and built the station pictured above and was close... one full set but not quite enough to give clear detail in places... like background folding into the model.... so started messing with different backgrounds and markings and took two sets ( one low angle, one higher) that I considered good as possible and tried processing those ... anyone using a proper camera might have better results, but the camera on the Sony Z5 is rated also.... and it also gave mixed results:

As mentioned the main aim is to get as many photos accepted in the sparse cloud... so all from one side and none from the other will not do etc. On sparse cloud generation presets you can set the level of effort from fast to exhaustive, and the time to process increases at each level. So to cut a long story short I had the windows tablet running occupied 12hrs a day for several days with the various efforts.

With the last set I had eventually started "exhaustive" to try to get it accepted, but after hours and still processing I figured that would not do either and decided to mess with some advanced controls to teach it all a lesson... as the progress bar neared complete I was ready for another half finished offering and had already thought of taking another set, or of using a different model maybe.... but there it came through with just 3 of the pictures left out. Those settings took under an hour for a good sparse cloud - acceptable.

The dense point cloud always takes longer, and on my computer it warned everytime of low RAM before starting, and that it might not finish.... but it always did.... took something over an hour.... the mesh is faster....and the textured mesh also


if you pick an intense setting, or if you adjust parameters to high definition.... etc.... you might end up with several hours computing for any step ... and when that is to produce a starting sparse cloud that you just have to wipe clean and start all over again from.....

Add to that that high definition settings do not necessarily provide better results either.


The Sparse Cloud settings that accepted all photos with 1 hr processing time, using title abbr. of first letter of word for all on "Advanced"


The Dense Cloud settings that gave minimal process time


Those are a starting point maybe, for better quality - higher settings work but increase processing time.

So https://sketchfab.com/models/99a18b9550 ... d905dcbc76
took twelve hours on my setup for dense cloud.

https://sketchfab.com/models/b0a5bc146a ... ec57266287 is the result of using the faster half hour dense cloud settings.

https://sketchfab.com/models/2412bd55a7 ... 515d434a00 is the faster settings plus an added step of Photoconsistency Based Optimization (PBO)

PBO only worked using lower resolution settings on my setup due to RAM size ( had to lower Res. and No. of Cameras etc.)

No effort made to tidy the results.

Mesh settings used on all were :

PBO-off (it can be run alone later - took half an hour in this test)

Took under quarter hr

Textured Mesh settings were Advanced at default.

Took around half an hour

Equipment used :

PC Intel Cherry Trail 1.5ghz @ 2gb Ram 10GB free on disk Win 10 1709
Camera Sony Z5.

Here is a good overview of the process

https://blog.sketchfab.com/getting-star ... phyr-free/

3Dflow has a series of tutorials

https://www.3dflow.net/technology/docum ... tutorials/

Of which Parameters
https://www.3dflow.net/3df-zephyr-param ... ing-guide/

Point cloud filtering
https://www.3dflow.net/tutorial-point-c ... h-filters/

https://www.3dflow.net/technology/docum ... d-quality/

though a bit technical, were of much use.

Also a good technical look at photo quality vs hardware at
http://www.clarkvision.com/articles/doe ... ze.matter/

I leave it at that, a basic intro with some tips to getting started and working ... the rest is anyone's own adventure.

You do not have the required permissions to view the files attached to this post.

Post Reply

Return to “Tips & DIY”