köln | süd

ein aus vier einzelbildern zusammengesetztes foto aus 50 m höhe. links das kaiser wilhelm gymnasium. ganz oben, im anschnitt, st. georg. darunter die einsturzstelle des stadtarchivs (wo heute einige stunden lang ein riesenkran tätig war). darunter die beiden neubauten, vom düsseldorfer bauherrn bpd pseudokölsch genannt “vielvrings” (angeblich kostet eine der sieben reihenhauseinheiten an die 1 mio €). und rechts die LVR-severinschule.

(c) ms/dpa
von unten

wikipedia | traveling

nach einer schier unendlich aufwändigen, angeblich jahrelangen vorarbeit der → wikimedia deutschland e. v.* sollten und konnten am wochenende einige wikipedia-autorInnen und -fotografInnen schlösser, türme, parks in berlin grunewald, babelsberg und potsdam von innen und außen und oben ablichten. ich habe nur den ersten tag mitgemacht, der für mich persönlich auf ungefähr 10 panoramafotos im → orangerieschloss von sanssouci hinauslief. wir hatten da privilegierten zugang und eine kuratierte führung.

kam mir bisschen viel aufwand vor, aber es war gut, da gewesen zu sein.

rückfahrt von gleis 5

* das ist der verein, an den ein großteil der jährlichen deutschen spenden an die wikipedia geht; die wikimedia hat nichts mit inhalten zu tun, den wikipedia-autorInnen seit über 15 jahren unbezahlt erstellen – auch wenn die dort fest angestellten mitarbeiterInnen manchmal meinen, sie wären der kern der bewegung.

Am.a.z | on

diese spam-mail ist gefährlich. sie fängt mit dem schein-absender “Versandstatus” an, gefolgt von einer schlecht getarnten dumm-adresse.net. das allein sollte genügen, sie zu löschen.

es geht weiter mit dem betreff einer amazon-überraschung, die von leerzeichen und punkten nur so wimmelt. der grund dafür ist, dass der betreff ohne diesen unfug in jedem spamfilter hängen bliebe. so kam die mail immerhin zu mir durch.

der eigentliche text der nachricht besteht aus der in blau gehaltenen anrede: sehr geehrter max.schoenherr. blau, weil er mit einem link hinterlegt ist. der harmlos scheinende klick auf das “sehr geehrte…” führt ins verderben, nämlich auf eine webseite, deren adresse ich hier mal in ihre ganzen schönheit in rot aufgeschrieben habe.

darf ich noch einen tipp geben? → hier ist, ein halbes jahr zum nachhören, mein 14-minütiger bericht über die IT-sicherheitskonferenz in london, wo der mensch als schwachstelle behandelt wurde. die in windows und mac eingebaute firewall und die kostenlose oder bezahlte antivirus-software fängt 90% der angriffe ab, was bedeutet: 10 % kommen durch. nur der anwender, der die mail sieht und kritisch beäugt, wird mit den 10% fertig. wer generell zu schnell klickt, hat schon verloren. die meisten privaten PCs dürften vermutlich infiziert sein, ohne dass es die benutzer wissen. heutige viren haben höheres im sinn, als dem user einen schnellen schreck einzujagen.

bus | stop

lots of cameras

today i waited for the bus. not a real wait, maybe 4 minutes. the rain had just stopped, i hid my umbrella behind a fence and took about 30 quick photos with my smartphone. you see the position of my camera in the screenshot above.

the bus came sooner than expected. i thought my 30 rough photos were certainly not enough for creating a 3D scene from them. in fact, they were!

a proper 3D model of the scene

i used autodesk recap photo and maya 2019 for this. both apps are free for non commercial usage.

spire | 3D

one of my favourite recording devices is the iZotope spire studio.

spire studio; looks strange. ain’t real anyway

in this experiment i took a few dozen photos around it, on a black table, loaded the images into an application which extracts 3D data from the photographs and creates proper 3D models of the device.

below are several views of the (incomplete) geometry.

spire recap 1
spire recap 2
spire recap 3
spire recap 4

abstrakt | from 3D

(c) ms

this abstract image was created from a series of trivial smartphone photographs in a radio studio. i shot the photos like for a panorama by just rotating myself and pointing the camera up or down. then i fed the 49 photos into autodesk’s recap photo app and let it create a 3D model from it. since the photo session was improvised in 2 minutes or so, and some areas in the studio were very dark and without contrast, the 3D object was lacking precision, large parts were just ignored. the advantage, however, is that in a 3D scene you can navigate as much as you please. the abstract art above shows a view from the top, with just a few round dots added later in photoshop.

this is the way how recap interpreted the photos and assembled the 3D scene. the model is upside down, which is quite strange. probably because i took the photos not in landscape mode.

below you find a couple of predecessors of the rendered scene.

brompton | 3D 2D 3D

today i placed my brompton bike in its folded position into a living room. with my DSLR camera i shot 130 photos from different angles. see the image above.

i then loaded the images into autodesk reCap Photo, a program for creating 3D objects from ordinary photographs. the last experiment with a telephone went well, so i wanted to try the same process with a much more complex object.

the result was sort of a shock: instead on concentrating on the folded bike, recap had measured the whole room:

autodesk recap interpretation of 130 photos

the scene looks so real because the textures were stitched in high resolution. the photo below is no photo, it’s a 3D computer scene. the black elements are parts of the wall, seen from the outside.

[all images on this page are (c) copyrighted!]


i exported the 3D scene and loaded it into autodesk maya. it arrives with yet another surprise: we see lots of camera icons, 130 in total!

here’s a closer look at the brompton. the whole scene consists of about 5 million polygons:

once the scene is in a 3D app such as maya you are free to move around. here is a view from the top:


back in recap photo you can pre-render the scene as xray:

or in a wireframe mode; it looks almost like a pencil sketch:

using arnold as the default maya render engine, you can get results like these:


and with the scene reduced to the geometry of the bike, you can build your own environments, for example by introducing a smooth light and a reflective surface for the bike:

here you see the problems: the spokes on the rear wheel are missing, the spokes on the front wheel are visible, but they are pieces of texture, not real geometry.

from a distance like in the image above, the bike looks convincing. not so with a very close look:

computer animation, especially for gaming, is about mesh reduction. that means you don’t want so many polygons. in the screenshot below you see three bikes which look almost the same. but the left one consists of 700,000 polygons, the middle one of 140,000 and the right one of 9,000 polygon “faces”.

the wireframe visualisation tells the technical truth:

and when you remove the complex texture from the 3D structure, the poly count becomes even more obvious. in other words: you don’t need that much geometry when you have excellent textures.

if you want to get serious about modeling and animating a bicycle in 3D, the 3D scan is a good starting point. and even if you use the geometry only as a reference for modeling, you might actually appreciate the rich texture which comes with the scanning process.

i stick to the bike as it is for now. but i’ll give it a face lift ;-)

humboldt | revisited

jetzt in der wikipedia
jetzt in der wikipedia

ich hatte vor zwei jahren das kölner humboldt-gymnasium von oben fotografiert. weil es bei der geringen, dort erlaubten flughöhe unmöglich war, das ganze gelände auf ein bild zu bekommen, musste ich der drohne einen flugplan verpassen, den sie ordentlich ausführte. weil sich aber das licht während der 10 flugminuten drastisch änderte (harte schatten von der schräg stehenden abendsonne, dann wieder bewölkt mit völlig harmonisch-diffusem licht) waren die ergebnisse beim anschließenden zusammenrechnen so schlecht, dass sie für die wikipedia unbrauchbar waren. siehe → dieses klagelied.

dank verbesserter stitching-software gelang das nun, und zwar mit den 2 jahre alten fotos. entscheidend war, dass die software (in diesem fall pix4D) aus den metadaten der aufnahmen GPS- und zeitmarken herauszieht und deswegen schon vom prinzip her keine fotos an stellen legt, die geometrisch unsinnig sind, auch wenn sie optisch zu passen scheinen.

das bild oben enthält noch einige detailfehler, die man in der 8000 pixel breiten wikimedia commons-version suchen kann, aber die architektur ist völlig intakt, man kann die inspektion des dachs vornehmen, wenn man will.

inzwischen wird an der schule heftig gebaut. es sieht bald ganz und gar nicht mehr so aus, wie hier.

2D to 3D | telephone

i rendered this telephone in autodesk maya, my preferred 3D computer animation program. the phone needs a lot of cleaning up. the cable is only partly visible, there are artifacts all over the place, including that odd object in the air.

the process was this: i put the real phone on a stool which i could rotate. i shot 129 images from different angles.

129 images

then i fed the photographs into the autodesk app “recap photo” and got this result:

3D model made from photographs
side view
rear view
rotation, by autodesk recap

recap offers several export options, for example obj. here is a rendering using a standard shader in maya, the same as in the image at the very top.

rendering in arnold with two light sources

a so called “playblast” of the maya scene

in maya, when you apply the complex texture image which comes with the 2D/3D process, the mapping is perfect, but details are missing:

rendering in arnold, textured fbx

finally, i exported the 3D data as STL, which is compatible with wikipedia. you find it there now.

ortho | photos

i’m familiar with drone photography, but not with aerial encounters which involve a lot of postprocessing, especially stitching of several images. the photo below was stitched together from 8 single, overlapping orthogonal drone shots and is almost perfect:

photo: ms/picture alliance

the whole open air project took about 15 minutes and produced 58 images. i had set up the flight pattern before actually taking off, using the swiss app drone harmony.

drone harmony pre visualisation

right before take-off the app asked me about the speed of the flight from one stop to the next, and off the mavic 2 pro went. in the brief video you see the blue sky (with the drone, somewhere in the distance) and the drone harmony app at work. it shows how the drone moves from one position to the next, stops, takes a photo and moves on.

the resulting 4k images were excellent. however, no stitching software could handle them. photoshop for example stitched certain parts of the landscape and ignored others.

well done, but only 8 out of 58 images.

there were obvious and subtle stitching failures like in the following image where the street was cut in two:

stitching error

the reason seems to be that objects on the ground look differently from shot to shot. example of two shots next to each other:

things look different from shot to shot.

the programmers of drone harmony advised me to look for professional drone photo stitching companies. and they explained that the change of perspective from photo to photo actually is a plus, because it provides 3D information. i then created a free short term account with the swiss company pix4D.com, uploaded the complete set of 58 photos and within 20 minutes got the – quite amazing result:

a 3D look from 58 flat images
3D navigation

the 3D visualisation came with a rich technical documentation. here’s my brief video showing the result:

interactive UI on mobile device

pix4D were poor in terms of service. i send an email with a question or two and got no reply.

autodesk, a software giant, has a 2D to 3D tool in their portofolio. it’s called Recap, or, if you’re dealing with photographs: Recap Photo. the results below are from recap. they come with a much less information about the scene than pix4D, they’re not getting the full picture of my drone flight either, but they have nice visualisation options. below are two examples of my garden drone project, computed by recap.

recap’s 3D interpretation
artifacts in the sky
DEM map of the area

wheat | bread

die bildstrecke beginnt 24 stunden nach dem ansetzen der ersten beiden sauerteig-stufen und endet mit dem fertigen brot. weizen backe ich gern sauer mit hefe.

weizenteig möchte geknetet werden

nach der teigruhe und dem falten

backen mit klimagaren

eine stunde später

lockere struktur und perfekte kruste

fertig

single | magic

one of, if not THE most innovative companies in computer graphics these days is allegorithmic from paris. here i’m using the 0.6 beta version of their new product “substance alchemist”. the three renderings, done in real time, with just ten or so clicks, were created from the photo above. click on the images to enlarge them.