Appearing as an enhanced version of the "Clone" feature present in most photo editing applications, the system analyzes the context of the image and not just the content of the adjacent pixels.
With a database of thousands of photos captured in different situations, the artificial intelligence system developed by NVIDIA can express the digital equivalent of imagination, realistically reconstructing the missing portions of the image. Practically, the system first determines the specificity of the picture and the objects in the field of vision, then approximates the missing portions
For example, after identifying a portion as representing a person's portrait, the AI system will rebuild face elements such as nose and eyes, even if they are almost completely missing in the image provided for analysis. Without the help of AI technologies, an image editing application can at most multiply the content of the adjacent portions, resulting in a chaotic blend of shades and textures
The "partial convolution" approach guarantees a higher level of coherence of the reconstructed image, even if the AI system does not "fully" understand what it is. However, using borrowed items from other pictures previously analyzed can yield unexpected results. For example, facial features reconstituted in a photo of missing portions too large to fix enough markers are roughly approximated, with the final result showing perhaps as credible, but different from the original image