Samsung Scientists develops AI which generates 3D video scenes. News has it this week that researchers with the Samsung Artificial Intelligence Center in Skolkovo and in the Moscow Institute of Science and Technology have published a new article about photorealistic scenes from a video. Last week, we saw in the news where tech teams where saying that Artificial Intelligence should be recognized as an Inventor.
The paper has details of the creation of a software that generates 3D animated heads from a single still image. This article is for researchers looking for ai generated graphics and ai generated video contents.
The news in details has it that 3 researchers at Samsung have created an AI which can generate realistic 3D renders of video scenes in movies.
In the published research detailing the neural network behind the artificial intelligence (AI), the 3 researchers explained the inefficient process of creating virtual scenes. A statement reads as follows;
- VMware & Google Cloud completes deal, as hybrid partner for hyperscalers
- Google Launches Anthos Hybrid Cloud Platform & Workload Management
- IBM Finalizes $34Billion Red Hat deal, vaults into multi-cloud
- Microsoft Launches Azure Security Center for IoT
“Creating virtual models of real scenes usually involves a lengthy pipeline of operations”. As a matter of fact, such modeling usually begins with a simple scanning process, where the raw scene geometry is captured using depth scanners or dense stereo matching and the photo-metric properties are captured using camera images.
If you listen closely, you will notice that the latter process usually provides noisy and incomplete point cloud that needs to be further processed by applying meshing approaches and certain surface reconstruction. Because of the mesh, the material estimation processes and texturing determine the photometric properties of surface fragments. After that, it stores them in the form of 2D parameterized maps, such as view-dependent textures, bump maps, surface light-fields and texture maps.
The final result will generate photo-realistic views of the modeled scene involves computationally-heavy rendering process such as radiance transfer estimation and/or ray tracing.”
To be more specific, a video input is converted into points which represent the geometry of the scene. Going forward, these geometry points are then rendered into computer graphics using a neural network, widely speeding up the process of rendering a photorealistic 3D scene in videos.
The result in a video of a 3D scene created by the AI is available online
Believe you me, this type of solution could one-day help game creation and development, most especially video game counterparts of different type of movies that are already being filmed.
Furthermore, by using this new technology, footage from a film set could provide a replica 3D environment for computer and android game developers to create interactive experiences in. Or, maybe, anyone could relive events like a wedding day using just a VR headset and an old video. Recently, Gartner says AI will push business value by supporting human decisions.
The 3 researchers explained that before such a point is reached, some improvements still need to be made to the program. For now, current scenes cannot be edited or altered and any large deviations from the original viewpoint results in artifacts.
But still, it’s a fascinating early insight at what could be possible in a not-so-distant future, as researched by the scientists. Finally, we focused on the topic titled; Samsung scientists create AI which generates realistic 3D renders of video scenes.