Making Of
Making Of
News
Videos
Software
Maya
Videos
PhotoScan is a photogrammetry software developed by Russia-based Agisoft that allows you to generate point clouds data, 3D models and their corresponding textures from a series of photos. It's an easy to use and affordable alternative to laser scan solutions.
I'll be reviewing the software from the point of view of a 3D/VFX artist. Let's see what PhotoScan has to offer.
In order to create 3D models the first thing you must do is to import a number of images from the object or scene into PhotoScan. The software supports the following file formats: .jpg, .tif, .png, .bmp, .exr, .pgm, .ppm, .mpo and .seq. The basic workflow is pretty straightforward:
Each step offers you to change the base settings up to your needs. The more photos you use, the more detailed your mesh will be. Also, an important aspect to keep in mind is that your computer resources will influence the complexity of the mesh you can generate. As it's an offline process, the more RAM you have the more polygons you'll be able to get in your object. The results will be even better if you have computers connected together via a network.
Even though the workflow is rather simple, one must learn a few tips for shooting the subject one wants to acquire in order to get quality results. It's easy to get frustrated otherwise, so make sure you read the shooting tips and the help file before using the software. You'll avoid bad results and frustration, something I experienced when I first used PhotoScan a few years ago.
Let's go over the different tools offered by the software through a real project: the acquisition and reconstruction of a tree trunk. For this project I took 33 pictures of the trunk shot all around its base. After the alignment process you can see blue rectangular shapes representing your camera position in 3D space, and the initial point cloud. Also, a bounding box appears where there is a bigger gathering of points in the cloud. The next steps of the process will be using that area only. You can resize and rotate it to include your mesh. Therefore all the surrounding trees, leaves and objects will be ignored.
The mesh can be generated from the point cloud that we just created, if it lacks detail it means you don't have enough data to be used in the mesh generation process. If that is the case, you can choose to use the next step in the process: the generation of a dense point cloud. This will be the longest step for each of your scans, indeed, for the tree trunk project it took approximately 6 hours on a i7 2600K @3.40GHz with 16Gb RAM.
Once this is done you can choose between sparse and dense cloud options for mesh building. If you find that there are some parts of the mesh that you won't need, you can easily remove them with the rectangle, circle or lasso tools. The gradual selection option lets you select polygons by their size or by connected component size. If your mesh isn't perfect and has tiny holes, there is also an option to close them, controlled by a slider.
A few words on the quality of the meshes generated by Photoscan: they usually have a great amount of detail and are highly accurate with respect to the original model. But this only holds true as long as you have done a good photo shoot. There are some exceptions: shooting inanimate models (trees, rocks, objects, etc.) is quite easy and can be done using a single camera. If you want to shoot a live model, however, like a human body, face, animal, etc., you'll need a camera rig that takes all the pictures at once. Why? Simply because live models breathe and have subtle motions, so no matter how fast you are at shooting photos with a single camera, your photos will not be consistent and the generated mesh will end up having 3D noise.
Depending on your needs you can use the decimation option to drastically reduce the amount of polygons while keeping the shape of your mesh. This is welcome when you need to do a retopology process in an external 3D application.
Like the other processes, creating the texture is mostly automated. All you need is to set the texture size, choose a mapping mode (the "Generic" option gives good results 90% of the time) and hit OK. That's it, we have a realistic CG tree trunk from only 33 pictures.
Maybe you're not a tree lover like me and would prefer to use the software to scan really huge areas from a plane. This is possible by switching from the Arbitrary to the Height Field Surface type in the mesh building process. With this method, you'll get an elevated terrain geometry as a result.
Generating 3D scans is an interesting process in itself, but the output is usually meant to be used in external applications. For this, one has several choices for exporting meshes:
In the File menu:
There's even an option to upload your model to Sketchfab or Verold.
In the Tools menu:
PhotoScan isn't a simple click and wait software. It offers plenty of useful options to help you get the best out of it.
Camera Calibration recognizes your camera model, the (electronic) lens connected to it and its focal length. If this doesn't work as expected, you can adjust the type of lens format: frame, fisheye, spherical, pixel size and so on.
You can use mask tools in your pictures to mask out unwanted elements. The batch process option allows you to launch all the steps of a basic scan process instead of launching each one after the other, a real time saver when you're scanning the same kind of objects shot in the same light conditions.
In case you're scanning a single person's face with many facial expressions, you can load your pictures into "chunks", which are processed separately, while keeping the generated meshes aligned to each other. PhotoScan offers you to align your chunks using 3 methods:
In the above list, I sorted these methods from the less precise to the most precise results. The point based method works already like a charm if you prepared your shooting session appropriately. As I don't own a camera rig and a photo studio, it's the solution I use most of the time. Did you shoot your model in more than 1 photo session? Don't worry, PhotoScan can also merge chunks, and guess what? The batch process also works with chunks.
The preferences of the program let you chose several languages for the software's UI. In the OpenCL tab you can set the number of Active CPU cores to speed up the processing, and believe me it really does speed up the process. The Network tab is here to set up the network processing.
If you don't own a top of the line machine to generate your models you can connect multiple machines together and go for the network processing solution (Pro version only). This can help speed-up processing greatly. However, be warned that it's not an easy task, and will require a little time and effort if you're a common CG artist with limited computer knowledge. Indeed there are actually no applications or shortcuts to launch servers and nodes. You have to do these steps via command lines and that is not really convenient in my opinion. I hope Agisoft will improve that in the next releases. I have to mention though, that I got quick and great support from the company through some mail exchanges and by visiting their forum.
Examples of work created with PhotoScan by Nicolas Brunet
Though it is not covered in this review, it should be noted that PhotoScan provides 4D videogrammetry capabilities (Pro version). That is, the possibility of generating 3D meshes from video.
In my opinion PhotoScan is one of those software you need to have in your toolset if you are a professional 3D artist. Sometimes you just don't have the time it takes to model a complex and realistic mesh. You never know when a client is going to ask for an archaeological excavation site to be modeled in 3D, or if an architect out of the blue may ask you to model the extension of a house stuck to a rock wall, and may remark that it has to look as realistic as possible. For a project I worked on in the past, I had to model rocks for a CG commercial. PhotoScan was a time saver, I could generate believable rocks and tweak them in an external application (3DS Max) to suit the mood of the project.
PhotoScan can also be used for various other aspects of 3D creation besides modeling: it can help create realistic textures (with some post work) such as believable displacement and normal maps, which you simply can't obtain by using software that transforms your pictures into approximate normal, bump and height maps. Besides the Pro version, PhotoScan is also available in an affordable standard version that provides most of the functionality you'll need as a CG artist.
With all these pros it's hard for me to find cons for the software. Of course other photogrammetry solutions are available. Some offer to generate your mesh online, and this way you don't need to own a powerful computer for the process. However, thanks to the fact that PhotoScan is an offline solution your projects will remain confidential, and you'll have total control in the creation process. That is something I like over some of its competitors. Another plus is that the software is frequently updated and it's already being used by well known video games and VFX studios.
Software
Reviews
Interviews
Terms and Conditions | Privacy Policy | Cookie Policy
Copyright ©2000-2024 CGPress. All rights reserved.