1280x735

2e52e3183df1e9661....jpg

๐Ÿงต Trying to model from few photos - fSpy/Blender techniques

Anonymous No. 848517

Anons, I'm trying - and failing - to model the tank in the image. The vehicle is the Tamoyo 1 from Bernardini. I'm using fSpy Camera Solver+Blender to basically do a poorman's photogrammetry. There are no good (1080, 4k) pictures of the vehicle, the one attached is basically one of the best images one can find.

What I'm doing is as follows:

1. Set axes references on a few different pictures. I think X and Y are generally fine since I can use the front and side of the vehicle, as I know they make 90 degrees to each other. Trying to use Z is more problematic as there are basically no Z references on the images.

2. Put the grid reference XYZ origin point on the same reference place in all the pictures.

3. Import the cameras into Blender, model and match from the different perspectives.

The problem I think is that the images are so bad the axes aren't properly aligned in fSpy, trying to set the 3 axes is especially hard.
Anyone willing to share some tips, perhaps even give a try on setting the axes? I can post more images if necessary. Ty anons.

600x350

mb3_tamoyo_l5.jpg

Anonymous No. 848534

look for overall dimensions and blueprints
is it this tank
https://en.wikipedia.org/wiki/Bernardini_MB-3_Tamoyo
Length 6.5 m (hull) [1]
Width 3.2 m[1]
Height 2.5 m [1]

2065x3081

ECYp2x6.jpg

Anonymous No. 848535

it says tamoyo 1

1280x924

MBT-Tamoyo-4.jpg

Anonymous No. 848536

much better

Anonymous No. 848537

>>848534
>>848535
>>848536
So.. The problem is that those drawings are not official blueprints. Well, the third one was made by the manufacturer, but it's more of a general sketch than a blueprint (you can check that many of the lines are crooked, the different views don't have the same proportions - front to side, side to back, front to back).

Currently I'm using the third image as reference to block out the vehicle, but I will still try to match it to actual pictures of the vehicle (3rd version in this case).

Anonymous No. 848539

>>848517
a somewhat basic tip: watch out for mismatched resolutions. if you're taking references from different sources, the import plugin will set the aspect ratio to that of the last imported camera, probably causing all others to shift.
also consider the fact that you won't always be able to pick the same visible origin on all references (for example, if you're tracing large buildings).

the way I do it is to first import one camera and do at least a rough blockout. then prepare second reference, pick some point common between the first and second ones, and put an empty there in your scene. parent everything else to that empty and clear its transforms - now the scene is ready for importing the second camera. after importing it, go back to the first camera and adjust its lens until it matches the new aspect ratio.
or you could expand all photos to the same canvas size. fwiw, back in 2.79 fspy could only reliably handle 16:9. or keep a separate scene for each camera, or whatever.

if you have about a dozen photos taken from the same location, just shove them into a photogrammetry solver. it takes less than you might think to get something workable.

620x411

MB-3_Tamoyo.jpg

Anonymous No. 848541

>>848539
I thought about using proper photogrammetry, but I don't have access to the vehicle nor a set of images good enough, in fact, photos are somewhat scarce, from different sources and with all sorts of other problems. I have searched for videos as well (for frame extracting), same thing, garbo quality from the 80s.

I wasn't aware of it changing aspect ratio, but I did see the cameras shifting. Thanks for pointing that out.
I think I can work common points between the images, but properly axes alignment is proving troublesome.

I'm using fSpy 1.0.3, Blender 2.92. No access to any camera info.

Anonymous No. 848544

You won't find any ACTUAL blueprints of 30-40 yo tanks or airplanes in the wild That stuff is treated as matter of national security. Design bureaus and arms manufacturers guard those like treasures.

720x366

1485968662_mb-3-t....jpg

Anonymous No. 848546

>>848544
I'm aware official drawings are hard to surface, hence I try to avoid the guessketches as much as possible.

Anonymous No. 848550

>>848537
second picture was photographed, you have to fix it in photoshop

1386x632

111.png

Anonymous No. 848948

Well, for now I decided to just model the Tamoyo 3 in anyway I can, just for practice. In the future I'll try to visit one of the surviving vehicles and get proper photos for photogrammetry.
Pic related is my attempt, so far. Next session I'll make some hull details.

472x456

tank turret.png

Anonymous No. 850469

Speaking of tanks, how do you keep all the surfaces flat when working with the part of the tank turret that involve more than 4 vertices going in different directions? I've tried using looptools flatten but that just causes other surfaces to go awry.

Anonymous No. 850478

>>848948
Looking good, but was the turret front armor that blocky?

Anonymous No. 850484

>>850469
huh?

Anonymous No. 850498

>>848948
You're doing really well so far. With some objects, you just have to remind yourself that you're making a model, not a replica. If you fudge the proportions a little bit, it's okay, especially if you capture all the details. I guarantee that if you took even a super high-end model kit, and scaled it up to be the same size as it's real-life counterpart, some parts would not be 1:1 accurate in their dimensions or angles.

Anonymous No. 850502

>>848517
just draw your own references based on pics and fill in the blanks with parts from other tanks or use your imagination, nobody will notice as long as you make it detailed enogh

1281x620

22.png

Anonymous No. 851297

>>850478
OP here, as far as the pictures go, yes, the version 3 was really blocky.

I recently bought a book on the vehicle which shows many pictures that are not on the internet. The manufacturer was trying to use composites and spaced armor in the turret and front hull, but had no previous experience with such materials, that's the reason it ended up being very blocky. The versions 1 & 2 had very angular facets, but no composites (and quite thin armor, max 70mm on v1 and v2 while v3 was up to 300mm).

>>850469
I think I got what you mean, the way I do is I create a big enough cube and use it as a boolean to cut the surface and guarantee a flat surface.

You can actually do this two ways: using two different objects and a boolean modifier, or you can go into edit mode and go into 'face->intersect (boolean)' with the cutter object selected.

Recent progress in pic (I don't use AA in Blender).

640x480

mb3_tamoyo_l4.jpg

Anonymous No. 851298

>>850478
Pic of v3 turret, to show the blockiness.

1920x807

FlatFaceMultiVert....png

Anonymous No. 851301

>>850469
I made pic to exemplify better:
1. I had a cube that I modified a little bit (to end up with a 5 vertices flat face). Suppose that's the turret you're working on.
2. I selected the bottom vertices (bright red) and used them to create a new cube
3. I used those vertices as a rotating point to get the angle I wanted.
4. Separated the selection into a new object
5. Boolean modifier -> difference

Anonymous No. 851304

>>850469
>>851301
Part 2:
What I would do in your case is create a cube through the common 'add cube' tool, then size it a little bit bigger than this face you need to flatten.
I would then 'extrude' the face(or slide vertices using shift V, non-clamp allows you to go beyond the current edges) of that particular face of the turret. In practice you're creating an 'extra' mass that you'll then cut, to guarantee the flatness. Finally, use the cube to flatten it.