Better camera for better textures

[RichardWren]

I recently took a quick scan of three berry scones I made with POP2.

https://skfb.ly/ozytH

For each scone I took top and bottom scans.

I exported the fused point clouds and the textured meshes that I obtained through Revoscan, and I imported everything into Cloudcompare for procesing.

The vertex color wasn’t great. The texures were slightly better but not very good either.

I had to handle all the meshes and the point clouds to align them in the excact possition, to later in ZBrush project the textures from the Revoscan mesh into the new mesh from Cloudcompare.

Because the texture was not good, I combined it with the colors from the point cloud to add a bit more color information.

I’m not sure if this is making any sense, or if this is really the best way of doing it, but the point is: The textures from the scanner aren’t good enough. If you take a look at the model I linked above, you will see.

I have a lightbox. The models were well illuminated. The scanner was at the recommended distance, but the images that the scanner captured look like they’re from an old webcam.

It seems that the camera in the POP2 (and maybe MINI too, didn’t check yet) are not very good, are they?

I would sugest for future products to invest in better cameras (and also better algorithms like Cloudcompare, but that’s for a whole other topic), so it’s easier to obtain scans with perfect quality.

I’m happy about how easy and quick it is to place the scones (or whatever object) into the rotating table, and in a few minutes getting a perfect scan. But this texture resolution issue is very upseting. I’m missing my old photogrammetry, where both mesh and textures were sharp -sigh-

I hope this sugestion helps to improve an already great product. I would gladly pay more for a version with a better camera

[PUTV]

I do agree … but

It is not so easy to add just beta camera resolution to the scanner .
If that was the case they would do that .
But you need to understand that each frame you capture create one photo , so 360 photos per one rotation or 4000 photos per full body scan .
In short 10 photos per second.

Now we go limitations with the data .
Since the devices are created to be portable and used with other devices than workstations you will run in problem to save so many photos in real time at higher resolution in so short time .

The files would be so huge that would be not usable on phones, tablets and smaller devices anymore due to lack of RAM to process it in real time .

You will need computer with higher level of RAM and VRAM to handle that , so not a simple laptop anymore would be usable.

They will need to update the chip so it could handle so many pictures at that resolution , then you will get the issue to save it quick on your workstation not to mention the device price would increase to less affordable price .

Most 7K scanners still use the same camera as POP2 .
As that is mostly recommend in this type of devices .

Now the problem is not really the camera resolution here but the texture mapping algorithms, in short the software .

When you use RGB color per vertex data make sure your point fusing and meshing is at the highest setting possible to keep as much points possible ( meshing in CC or Revo Studio) this will preserve most pixels since each point is a color pixel for that reason MINI has higher pixel rate than POP2 thanks to more dense point fusion .

For Texture Mapping in Revo Scan there is nothing you can do about , as the output is behind our control .

The only improvement I saw when scanning slightly on slower rotation using POP2

So what Revopoint need to do is working on improvement of the Texture Mapping in Revo Scan as that is the only problem here .
As replacing camera to higher resolution is not an option , and even if , you will still get bad textures and low RGB color per vertex data output regardless of the resolution as the amount of points controls the resolution .

High resolution RGB camera are usable mostly in scanners that use per frame/angle scanning .
That why photogrammetry provides so sharp results .

[RichardWren]

Thank you for the info, PUTV, now I understand the limitations.

It’s not very encouraging though. The way the camera is right now offers a very “cute” hobbist level for 3D models, but is not suitable for professional use. I can’t sell the scans I take if they have such blurry textures. The models look wonderful from a certain distance, but for clients who need closeups… it’s not gonna work

Would it be possible to make an attachment for a camera that grabs the body of the scanner in a known, permanent possition, takes high res images every X frames, and then uses those images in the Revoscan texturing process? Or to take less images per frame but in higher resolution?

Another idea would be to make a desktop scanner, that doesn’t need to work with a phone. The POP-Pro or something like that.

I absolutely love the scanner. Scanning is so much fun, I can’t believe they pay me to play like this all day, but the results really need to become sharper, or it will limit the userbase dramatically to people who don’t need textures.

[PUTV]

I agree , for my professional work I need to create albedo textures so this is not important , nobody want a photo as textures anyway with backed in lighting , not good for anybody that is interested in 3D rendering .

A proper textures for your 3D object need to include color Albedo ( no light no shadows baked in ) normal / bump maps etc … plus at least 4-8K resolution

Even if it had perfect 40 megapixels resolution I still have to remove all the lighting and shadows .

That why I prefer to make my own textures I shot with my DSLR so I can reach the standard , shot everything with polarized filters to get rid of any reflection

I use the current textures or RGB only as reference for texturing on top .

I don’t see it coming , not 22 years ago and not tomorrow for this price range or I would be out of job already

[RichardWren]

That’s a great idea. I will try that next time

[RichardWren]

Too much extra work though, that’s why it’d be nice if the camera took better photos to start with, then you only need to manipulate the textures to make a proper albedo, as I do when I scan through photogrammetry.

Let’s see if there’s anything Revopoint can do about that in the future.

[PUTV]

@RichardWren

Let’s see if there’s anything Revopoint can do about that in the future.

Hopefully they will . I do reminds them very often about that .

[RichardWren]

Today I would like to show a model I did using Revopoint Pop 2 versus a model I did with photogrammetry, to insist to our friends at Revopoint to take a look at the camera resolution/texture making and see if they can overhaul it somehow.

As PUTV already explained it, there are limitations that justify the bad resolution. But if you take a look at the example, you will see how important it is to work harder on this issue.

Here is the model: Japanese Green Pumpkins - Buy Royalty Free 3D model by SM5 by Heledahn (@heledahn) [8fbbea4] - Sketchfab Store

There are three pumpkins. The bigest one I scanned using photogrammetry (less than a hundred photos, if I remember correctly, and point cloud created in Meshroom). I scanned the other two with Pop 2.

I cleaned and remeshed all the models to have a very low poly count.

Now, if you compare the textures, you can apreciate that the photogrammetry pumpkin has so many details. This is especially evident when you look at the bottom. (I must add that the photos I used for the photogrammetry were not very good, and still the finished model looks quite good.)

The Pop 2 models are blurred. The details are only there because of the outstanding mesh captured by the Pop 2, but the textures are very, very, very bad.

Please, Revopoint team, see if there is anything you can do to fix this.

Taking a few images every X frames that are higher resolution, and texturing the model with them is an idea. Or allow the users to input photos taken with an external camera and somehow align them using feature matching to do the texturing. This is not my field, so I don’t know what can or can’t be done, but I insist that something needs to be improved.

Textures should be as important as the mesh. And I can assure anyone that the mesh quality of Pop 2 is first-class. Let’s see if we can match the texturing to the surface detail!

[PUTV]

@RichardWren

As PUTV already explained it, there are limitations that justify the bad resolution. But if you take a look at the example, you will see how important it is to work harder on this issue.

Hi Richard , limitation in 3D structured light portable 3D scanner technology, not by choice .

Artec Leo use 2.3 megapixel camera cutting edge technology at $24K
POP2 use 2 megapixels camera at $699

It is not gonna happen unless you want your PC to be attached to your scanner permanently for support or pay big price.

However texture mapping can be improved in Revo Scan as that is not the optimal quality here yet so a lot of things can be addressed to make it better to produce the optimal results even with 2 megapixels camera .

For better texture quality you need also scan the objects from all angles as one missed angle will already created blurred area on the textures automatic by the program resulting in fussy appearance .

Don’t get me wrong Richard , I would love to get better quality from the texture mapping in Revo Scan .
Proper white balance and shadow blending etc … it is not optimal at this moment and the output is never consistent .

And Photogrammetry is bad comparison , because if you want manually capture your photos to use it in mapping of your POP2 scans , then you can just keep doing photogrammetry, as you already made the half job , and don’t need to use POP2 at all, why? because all the pictures you need to capture for texture mapping will produce a mesh in your photogrammetry software with a better quality so why bother ?

So yes I am for better and improved texture mapping function in Revo Scan , I also want Texture Mapping in Revo Studio so I can generate textures from color scans I merged or edited with re-mapping function .
It is all possible to be improved and up to Revopoint if they want to address it in the future .

But better camera for better textures is a wrong call here as having 3 times better camera will still deliver the same results with the current textures mapping function in Revo Scan .

[RichardWren]

Thank you PUTV.

Better camera for better textures is the title I wrote a month ago before you explained the limitations. I didn’t see the need to open a new thread to insist to Revopoint of the need for better textures

This is something important that people like us, who need high quality textures, must strongly insist about, otherwise they won’t fix it, and that would be a pity.

The way things are now, I spend two hours, give or take, editing the color points and the textures from my Pop 2 to achieve something far from decent.

I make models for an animation project, so the scans are going to be in the background and it wont be noticable if the textures are a bit blurred, but I don’t feel proud to sell said models professionally, seeing how bad the textures are.

It would be nice if they could implement a way of adding custom images (front, right side, left side, top, bottom, and back), and with their feature matching algorithm align them and project them onto the scan. Again, just a sugestion. But it would go a long way to improve the quality of the models.

Because as it is right now, the textured models are only good for hobbists or for background props that nobody is really going to see.

Otherwise is like you said: Why bother using Pop 2. Why bother buying the new Range?

It takes about an hour to set up an object and take a hundred photos for photogrammetry, feed them to the algorithm, and obtain a perfect scan with perfect textures, for free. With the 800$ one can even get a decent Canon.

But I like the Pop 2. I want to use it professionally. I hope Revopoint will hear us out and improve this problem. I wouldn’t like to have to go back to photogrammetry after all.

[PUTV]

@RichardWren

make models for an animation project, so the scans are going to be in the background and it wont be noticable if the textures are a bit blurred, but I don’t feel proud to sell said models professionally, seeing how bad the textures are.

I fully understand you Richard , that would be a game changer and I advocate for it since POP2 release …

@RichardWren

would be nice if they could implement a way of adding custom images (front, right side, left side, top, bottom, and back), and with their feature matching algorithm align them and project them onto the scan. Again, just a sugestion. But it would go a long way to improve the quality of the models.

I proposed it so many times already as I see it as only option to improve the quality of already sold devices . Plus better texture mapping quality .

@RichardWren

Why bother buying the new Range?

That’s the point , beside scanning auto parts there are the full body scans, rooms, furnitures etc that will need textures , the game breaker here , at that accuracy level without proper textures it not gonna be used by anybody but only the people that want meshes and nothing else .
Personally I don’t need textures at this level as it is not usable for my pro work , but if I could have it in better quality it could save me ton of work in the process even for second plane models and assets.

@RichardWren

takes about an hour to set up an object and take a hundred photos for photogrammetry, feed them to the algorithm, and obtain a perfect scan with perfect textures, for free. With the 800$ one can even get a decent Canon.

DSLR cameras are limited how much photos they can take before the end of their lifespan, so you would need 24 DSLR camera every day to make as much pictures that Range can produce a day in 24h. What is not really cheap on long run .
DSLR attached to Range doing pictures will die every hour of continuous work at the same speed and frames per sec .
For that reason you will never find this level of texture quality in portable light structured 3D scanner , technically not possible.

@RichardWren

I wouldn’t like to have to go back to photogrammetry after all.

Hopefully things improve … it almost was there with MINI but get lost

Trust me the 2 megapixels is capable to provide you with very sharp and derailed textures … but the software and texture mapping at this moment can’t .

This means the hope is still there and huge room for improvement.

[RichardWren]

Thank you, PUTV

I really hope they will listen.

[PUTV]

@RichardWren

really hope they will listen.

They always listen Richard but sometimes the task is difficult to be accomplished for one or another reason in short time.

Revopoint just started last year providing us with new goodies and it is in their best interest to improve as that is important to their brand after all .