So if you've ever tried photogrammetry or 3D scanning and were frustrated by the inability to capture the bottom side of your objects or get a closed mesh, you're not alone. There are lots of clever ways to approach this, like placing your object on

something thin or suspending it by wire, but this isn't always possible depending on the shape and weight of the object. Doing it this way is risky; it can make your object move, negatively impacting scan quality, and

requires a bit of mesh and texture cleanup in post. It's not very convenient. Now, fortunately, there's a simpler way that doesn't involve using sketchy rigs to hold your object upright. Now, full disclaimer: this video is sponsored by Capturing Reality, and we'll

be using RealityCapture software to generate our 3D model. So let me show you how to get a clean, closed mesh and how to scan the bottom side of any object. Now, first off, as always, the basics of photogrammetry apply. You can

watch my intro to photogrammetry video right here. I'll put the link below the like button. So let's grab the object we want to scan and set it down on a rock-solid surface. I do want to take a moment to talk about what kind

of objects you can scan using this method. It should go without saying that you have to pick an object that doesn't deform when you move it. It needs to be an object you can grab, rotate, flip onto its side, and that won't

Bend, move, or change the shape of a solid object. If you try to scan something like this, or anything soft that does deform, this method I'm about to show you will not work. The alignment process is going to fail; you're going to end up with a

complete mess. With that said, ensure flat, even lighting. Overcast days are perfect, or shoot in the shade. Always shoot raw photos, and if you're using a phone, do consider an app that will allow you to manually choose as many settings as possible. With these settings in mind, let's start

taking photos of our subject. We want to get photos from every possible angle of our model here, at various levels of elevation. So take photos around your object like this, while following the rough shape of it. The more photos you get of any single point, the better your

alignment process will be, and as a result, the cleaner the model. As always, make sure you have enough overlap between photos to ensure proper alignment later. 50 to 60% overlap is a good starting point. If you're in doubt, it's better to err on the side of

shooting too many photos than you think you need. Without enough overlap, you're going to have trouble aligning the top and bottom half of your object later, as you're going to see toward the end of the video. So stick around to see that. With your first loop done, increase your

elevation just a little bit while maintaining the same distance from the object, and perform yet another loop around it. Repeat this process until you're almost shooting straight down. You might even think you're taking too many photos, but again, it's way better to have too many

than not enough; you can always discard some later, but it's not always possible to reshoot, not to mention it's very inconvenient. So, with the top half of our object here scanned, let's flip it upside down and repeat the same process as earlier. Again, be sure

not to deform your object at all when flipping it, because bad things will happen. Once you've taken photos from both sides, we're now going to have two sets of images: one for the top half and one for the bottom half. Usually, I like to take one dark frame

between image sets so I can easily find which set is which later. With all of our photos taken, the hard part is done—seriously, the rest is easy! Next, let's bring our photos into Adobe Lightroom. I'm using Lightroom, but you can use

any other photo editing package that you want. Lightroom is just really great for batch editing photos, and we're going to be using Lightroom to brighten them a bit, so this is going to help the texturing process. Again, you'll see me go through this in depth in my intro to reality

capture video; right here, link below. Start by editing one of the photos. I'm simply going to lift the shadows, lower the highlights, and create a flatter image. Adjust the exposure a tiny little bit and make sure your white balance is okay. And because we didn't change our camera

settings during the shoot, we can sync this one photo correction to all of our photos, and everything should look the same. By clicking on the sync settings here, click OK, and now you'll see our edits have been applied to all of our images. Now, here's the important part.

We're going to export them in two batches, in separate folders. I'm going to select the images for the top half of my object here and export that into a folder named 'Top' by clicking on the export button right here and selecting my destination folder. We can export

these to JPEGs for now. The reason we wanted to shoot in RAW was to give us the latitude to edit and turn down those highlights and lift those shadows later, so exporting as JPEG is fine. Then I'm going to do the same for the

bottom half by selecting the photos of the bottom part of this log and exporting those into the 'Bottom' folder. Now let's open up RealityCapture, and we can drag and drop the entire 'Top' folder that we exported from Lightroom directly into our viewport in RealityCapture.

Let's hit the F6 key to align our images, and after waiting a bit, you're going to get your point cloud right here, and it looks great. Now let's just set our reconstruction region by clicking right up here and clicking on 'Set Reconstruction Region'. The

reason we do this is that RealityCapture will only process the area that we want. A handy tip here is to use the numpad keys to switch to orthographic views, which makes lining this up a lot easier. You can press 0 on the numpad to revert

back to perspective view. And now we are ready to generate the mesh by clicking on the 'Mesh Model' tab. We can simply start by creating a preview mesh, as this meshing process isn't really important for now, and you're going to see why real soon. Normal and

High mode is going to take way longer, so preview is fine for now. Once the mesh is generated, you'll see we only have the top part of this log. So now we need to generate what we call an object mask to combine the top and bottom parts of our object here. So let's

start off by cutting out any part that is not part of the object we're trying to scan. And in this case, it's the support beam that we have the log resting on right here. So by going to the Tools tab, Cut by Box, and then clicking on Set

Reconstruction Region, I'm going to go to the orthographic side view, and I'm going to chop out the bottom part here, clipping out any of the support beams. And for the record, this does not have to be perfect; don't worry. Then we can click on Cut Outer

to remove everything that is outside our reconstruction region from our scan. Perfect! With all the unwanted triangles gone, now we can generate our masks again. In the Tools tab, click on Depth and Mask and set where you want your mask to be exported to. We need to navigate to the

exact same top folder where these images are found in the dialog box that shows up here. The only thing to change is disable the Export Camera Depth; set this to 'No.' And that's it. Now, just a heads up for full transparency: if you're using RealityCapture's PPI setup, you'll need to use some PPI credits when exporting the mask, just like when you use credits to pay for exporting a model out of RealityCapture. However, once these images that you're using are licensed, so to speak, you will not need to pay for them again later when

You export your model; it's a bit of a bummer because I would love to have the option to at least see my final model combined and all before having to pay, but that's the way it is for now. I believe the devs are working on a

solution and a compromise here. So, with that done, let's go open the top folder here, and you're going to see we have perfect masks generated for every single image. This is absolutely awesome because manually painting masks for, you know, two or three hundred images is not fun.

So this mask is going to tell RealityCapture to only use the parts of the image that are white. It eliminates the background and allows us to merge the top and bottom halves together. With the top half done, now we can process the bottom half of this log

in RealityCapture. Let's select all of our images by pressing Ctrl+A, and you'll see all the cameras are selected. Now we can press Ctrl+R to disable them from further processing. You'll see they change color here; it's a handy visual indication that they are disabled.

Now let's import the images from the bottom half of the log by dragging and dropping the bottom folder directly into our viewport. You'll see here we now have 295 images, up from 167. Press F6 again to align these, and now

that these new images are aligned properly, you'll see we now have two components, which isn't quite what we want. To keep it tidy, I'm going to rename this component to 'Top' and rename the new component here to 'Bottom.' Now we know which one is which.

Now again, we can set our reconstruction region and then generate our mesh just like before, using the preview mode to quickly generate a mesh. With that mesh completed, let's use the 'Cut by Box' mode again in the Tools tab and cut outer again. This is going to

remove the support beams which we don't want. Same as before, next, let's generate the mask for this component by going to Tools, Depth, and Mask. Again, you need to export these masks to the bottom folder respectively, and RealityCapture will remember the settings that we

used earlier, so we don't need to do anything else. Now we have two batches of masks generated. Previously, in older versions of RealityCapture, we needed to have two separate projects to combine the two halves, but not anymore. Now we can simply drag and drop both the top and bottom folders of

our images here directly into the viewport. But you'll see the total amount of images did not change; it's still 295. But if we click on any of the images here and go to the Image 2D View tab and then click on Mask, you'll see every single image has its

mask loaded in automatically for us—it's magical! So, by changing my layout back to the one-plus-one mode, I'm going to select all of my images with Ctrl+A again and press Ctrl+R again to enable everything. And now, with that done, we can cross our

fingers, press F6 to align everything again, and if everything turned out great, you should get one perfect model with all of the cameras aligned perfectly. You get a fully closed mesh; everyone is happy with both sides matching in one whole, complete, closed object. But sometimes—a lot of times—

You'll get two or more new components. The two sets didn't align perfectly, and this is usually caused by a few different things: either you didn't take enough photos, the space you moved between photos was too large, some photos were blurry, noisy, or just too dark.

Getting this to align perfectly takes practice, but don't worry; we can still fix that. First, obviously you can go reshoot and take more or better photos, but reshooting isn't always an option, so we need to use what we call control points to help RealityCapture align those two components.

So I was initially going to narrate this part like I did the rest of the video, but this part can be a little tricky, so I figured I'd hold your hand through the process to make sure you don't miss anything. So now what we

need to do is assign control points to both of our components here in order to merge the top and bottom half of our object. So what I'm going to do here is go ahead and change the layout right here up top to give us a 2D image preview. Next, we're going to click on the 'Scene 3D Tools' tab and click on 'Control Point' right here.

You'll see that, as I move the mouse around, all these lines you're seeing are all of the cameras being aligned to that specific point on the point cloud. So we can use this to manually place control points in 3D space. Now, what's really important is that we need to assign a control point to something that is very easily recognizable and distinguishable across both

components, something that really stands out, a little spot that has high contrast that you can easily say, yep, that is indeed the correct point. Very important. So right here, I'm going to zoom in and you'll see I have this little piece of wood sticking out here. This is

perfect. This is exactly what I want. This is very obvious; it stands out, it's high contrast, easy to recognize. I should be able to find this pretty easily on my bottom component. So I'm going to go ahead and left-click right here, and don't worry, that spot doesn't need to be

perfect right now. I'm going to right-click now to get out of control point mode, click on my first image right here on the left-hand side, over here, and in the 2D view mode, I'm going to zoom in here and I'm going to move this control

point by clicking and dragging on it right over here because I know for a fact, all my other images, this spot stands out a little bit more. Now again, I can click and hold on it and press the down arrow key to go to my next image. Again, I'm still holding the left click.

Here, I'm going to move the control point to that same spot right here. Let go, and now you'll see that control point has moved, and as a result, we now have way more cameras being aligned to that specific spot. It was indeed a better control point spot.

So now again, I'm going to click on the top one here, click again in my 2D view, press the F key to zoom to that control point spot. Now we need to go ahead and visually confirm all of these images. You see all these images have a nice little green

Plus, next to it, by clicking on the image here. We can go confirm that; yep, that's indeed the right control point. We can confirm it. There are two ways to do this: there's the simple way, by clicking on the little green confirm button right

here on the left-hand side, or an easier way to do that is to click the next image in our 2D view, press the F key to zoom to that spot, zoom in, click and hold on the control point, and again press the down arrow key

to confirm it. It's automatically going to bring you to the next image, and we can just press the down arrow key over and over like this. And, as you'll see, it is cycling through the images to confirm that, yes, indeed, this is the correct control point. Do

this for all images that were aligned. And now you'll see, on the left-hand side, all of these images have been aligned correctly. We've now assigned our first control point on our top component. Now we need to go ahead and assign the same control point to our bottom component, so

in order to do that, we're going to go on the left-hand side, scroll down, and we're going to select 'Component 1' because this is our new bottom component. Click on this, and now we need to find that same little piece of wood sticking out on the bottom

component. We need to find that same spot and assign a control point to that same one. We are now in the process of telling RealityCapture which point is which across both components, and then, as a result, it will be able to merge the two together. So what we're going to do again

So we're going to go to the Scene 3D Tools tab up here again, click on Control Points, but this time we don't want to just click here because if we do that, it's going to create a new control point, and we don't want that. We want to assign the

same control point to this component. So we need to go ahead and click on Point 0 right here, and then we can go find our point in common. Now you'll see right here, it's a little bit hard to see, but that is indeed our little piece of

wood, our little stick that we saw in our previous component. So I'm going to go ahead and left-click on this right here, and now you'll see we have a whole bunch of images that were aligned correctly. Now we need to go confirm those by clicking on them right here

and zooming in. I'm going to confirm that it is indeed, yep, at the right place. I'm going to click and hold the control point, press the down arrow key again, and as a result, you'll see it confirmed for you on the left-hand side.

Now I do want to take a moment to talk about why RealityCapture had a hard time aligning these images to begin with, and that's because you'll see right here where I place my control point, this little twig here is completely out of focus. My depth of field was too shallow.

As you can see, if I zoom out, I focused on the front of the log here, but because I was so close, even though I was shooting with an aperture of f/10, I didn't have enough of the shot in focus here, and so that's why Reality

Capture may have had a hard time aligning this initially; that's why it's very important to get a clean data set to begin with, to avoid alignment issues like this. Now, RealityCapture does know that it should focus its attention on what's in the center of the object and

less so on the edges, but still having more of your object in sharp focus is absolutely very helpful. So now, as you can see, we have our first control point assigned to both of our components. Now we need to repeat this process across two or more control points.

We need to find yet another point in common across both components in order for RealityCapture to merge those components together. Two is the absolute minimum; sometimes three or more is necessary. Usually, in my experience, two or three has worked fine, but again, depending on your data set, you may need

more control points. So what we're going to do now is we're going to find yet another point in common and assign a control point there. So again, I'm going to go to the Tools tab up here, click on Control Points, and try to find something else in common

across our images here. I'm liking this area right here; this seems like a pretty good spot, easily recognizable. It will be very easy to find on my other component. Zooming in, I like this little point right here; it should be very easy to find. So I'm going to click on that right

here, and as a result, you'll see on the left-hand side we have a whole bunch of images that have been aligned. So I'm going to right-click to get out of Control Point mode, click on an image here on the left-hand side, and zoom in.

Click and hold on the control point and press the down arrow key to confirm it and proceed to the next image. Cycle through these images until you've confirmed all of them. Now, with the control point added to our bottom component, we need to go here on the left

hand side, select our top component, which is component 0, and assign our second control point to that same spot on the top component. So again, going to the Scene 3D tool tab up here, click on Control Points, and of course, again make sure I click on point 1 right here

otherwise, again, it's going to create a new control point, and we don't want that. Find that same spot and, in this case, this little thing right here, and left-click to add a control point. You'll see, as a result, we have a whole bunch more images that have been

aligned successfully. Right-click to get out of control point mode. Click on the first image on the left here, and in the 2D view, you'll see that my 2D view is selected because I have a blue border here. Press the F key to zoom in 2.1

and click and hold and press the down arrow key to confirm it. Do the same for all of these images as well, and just like that, we now have two control points assigned to both of our components, the top and the bottom one. You may need to repeat this process for

a third or more control points, but that totally depends on your scan. Now, I do want to take the time to show you another way that we can assign control points because sometimes when we're using the control points tool, it can be a little bit hard to

Really accurately add this control point like this. Another simple way of doing this: we're going to manually create a new control point right here on the left-hand side. So now we have control point two, and now, using the 2D view here, I'm going to zoom

in, and let's say I want to add my control point to this thing right here. I can click and drag control point two onto that point right here, using the 2D view. Drag and drop like this, and now, using the arrow keys to toggle between

these images, I can drag and drop point two on that same spot on another image, and you'll see, just like that, we have a whole bunch of these cameras aligned to this point. As I toggle through the images, you'll see these cameras were able to find this point correctly. So I

thought I'd show you just another way to assign control points. Now we can attempt to merge these two components by going to the Alignment tab up top here and clicking on "Merge Components". If everything worked out correctly, we can see all of our cameras in a dome

around our log here, and we have the top and bottom half of our log combined as one seamless, closed object. If, however, the "Merge Components" didn't work, you'll see you have new components created here. You can safely delete those and try again with our previous components. Find more common areas

between the components and assign more control points that way. As you can see, this can be a very slow and tedious process, and it's why making sure your images aren't too blurry, or making sure you have enough depth of field to work with, and your images are well-exposed, not too

Noisy. Taking the time to shoot correctly may seem boring and dull, but I can assure you that taking the time to get a clean data set will save you a lot of time in the long run. It takes practice, but eventually you're going to start scanning objects and

getting the alignment perfect on the first try, without the need for more control points. Now we can finally go to the mesh model tab and generate our final model. If you're going for the absolute best quality, go with 'high'. High does take way, way longer to make, though, so keep that

in mind. Often, normal reconstruction is good enough, but I thought you might want to know that there is a pretty big difference. With that done, now we have a completed, fully closed model that we can see and use from all angles. Guys, I know this is a lot to take in,

but it's a very powerful tool to have under your belt because close 3D meshes, as you well know, are infinitely more useful to us 3D artists. So, just a bit of an announcement: I wanted to let you know that I've just

started my Patreon page. So, if you're ever interested in supporting this channel, go check out my Patreon link down below. As a patron, you'll get plenty of 3D models that I create using photogrammetry moving forward. So guys, I hope you found this video helpful. If you

did, do consider subscribing and giving it the old thumbs up. And happy rendering!

How to 3D Scan All Sides of Objects - RealityCapture

How to 3D Scan All Sides of Objects - RealityCapture

Scripts