Autodesk ReCap 360
Data acquisition has become an integral part of the visual-effects process. Data might come from a LIDAR scanner, such as something from Faro; or a structured light scanner from Artec or Mantis, or via a Kinect. It may be derived from footage, like that from a drone, or it may be from a series of photographs from a DSLR.
And all that requires the efficient processing of that data.
Autodesk has a product called ReCap 360, which focuses on the importing of point cloud data — primary acquired from LIDAR scanners. Its critical functionality focuses on larger projects like construction and so forth, where accuracy is absolutely necessary. Frequently, scans are generated from multiple positions to prevent laser shadows (where the laser gets shadowed by a FG object). All these scans have to be registered to one another to make one cohesive scan.
ReCap has automatic registration algorithms to help with that — but also manual tools in case the automated one fails. This also allows for pinpointing very specific points in the world — on top of GPS tagging, for when your location is important to the job.
For us in the film and animation world, we may not need that level of control. But its great that it’s there, and it most definitely helps for tracking cameras in shots.
I happened to have some Faro data lying around from a recent project and was able to throw it into ReCap without any preparation at all. Drag and dropping the .FLS file into the window was all it took to get me on my way.
The user experience design for ReCap is refreshingly simple. I never had to pick up a manual once. All the tools are pretty self explanatory. And I was able to import the data, cull unnecessary points, and generally get the data set to a point where is was ready to go, all within the hour.
What I don’t particularly like is that some critical functions have been relegated to the cloud. This includes processing photos into a mesh, auto cleaning a point cloud of errant data (like people), and creating a mesh from the point cloud. If the server is down (like it was when I tried), or if your internet connection is out, then there isn’t an alternative. I also would like some consistency between ReCap and other Autodesk products like Maya and Max as far as mouse navigation goes — at least give me the option to switch to Maya nav.
Ultimately, easy to work with, simple interface. Great if you might be receiving scan data from multiple services —it’ll import quite a few scan formats. And it’ll export to quite a few point data formats.
Now, the sister to ReCap is a program called ReMake (previously known as Memento).
While ReCap is more about the importing of cloud data, Remake is more about processing the meshes that are created from that data — and then some.
So, if you had a point cloud, you could isolate an object in the cloud with in ReCap 360 — say, a car — and export it (through the cloud service) to a mesh. Retrieve it, and import it back into ReMake. However, LIDAR data is invariably dense. And you don’t want that flowing downstream to the animator or texture artists or 3D printer. So ReMake has both tools for decimating a mesh down to a target poly count and retopologizing tools.
ReMake also uses the same algorithm for extracting a mesh via photogrammetry that ReCap would use. In fact, Autodesk uses the same math in a number of different programs including 123D Catch, which is the consumer version that lets you generate meshes with photos from your phone. The difference with ReMake is that you can choose whether or not you want to process the photos locally, or you want to throw them up to the cloud for processing. If you happen to have an older computer or laptop without a robust GPU (greater than 2GB of video RAM), you make want to consider throwing it to Autodesk. Otherwise, it would make a good excuse to run and get a coffee.
For the 3D printing side of things, ReMake has a whole toolset for analyzing, assessing and repairing common problems with your mesh. So, you can clean the model and sanity check it before you print it out the model. There is a fantastic video on the website showing the fabrication of a prop gun, photographing it, building the mesh from the photos, loading the mesh into Maya and designing a whole gun around it, bringing it back into ReMake for a print check, printing the components of the gun, and then reassembling and painting the real prop. It’s frankly a bit inspirational.
The user experience design is similar to ReCap in that the interface is boiled down to just the essentials and the workflow feels natural.
Since I’m much more likely to have access to a DSLR than a Faro LIDAR Scanner, I think that I would lean toward ReMake as a more likely tool of choice.
And then there is a little company named CapturingReality, with its product RealityCapture, which appears to put all of the essentials into one robust package.
Reality Capture can derive meshes from LIDAR data or from photos — lots of photos, the more the better (which is the case with ReMake as well). But RC can get a more refined mesh when using both LIDAR and Photos. It has math going on to align the photos to the point cloud, and be able to refine the data set, including filling in holes where the laser may not have hit. Pretty fancy stuff.
And the processing is quite fast — multi-threaded on both CPU and GPU, which means the more robust your graphics card, the faster it’s going to run. And the acceleration is in Cuda, so NVidia is going to be a good way to go. The software has a draft mode to speed up processing further, and is designed to be used out in the field and on-set, where things are moving fast and furious and you need to verify that the data you’ve acquired is going to meet the needs of the project.
There are plenty of tools for refining and aligning the scans, which include DSM and geo referencing for those really big areas. And like ReCap 360, there are filters for getting rid of moving people
One of the alignment features I’m particularly fond of is how ReMake manually selects features in the photos that are seen from multiple angles, which in turn refines the positions of the cameras, which then recalculates to make the mesh more refined. Its a lot like camera tracking in a way — well, not in a way … in fact, it’s exactly camera tracking. But instead of a camera path, you are helping it triangulate the surface of the object in the photos.
The subscription cost may be a bit hefty for many outside of a studio, but CapturingReality does offer a three-month subscription package for 99 euros with the limitations being 2,500 photos per project and no tech support. But at that price smaller companies and freelancers can certainly benefit.