Intel introduced a revolutionary neural network for creating 360-degree panoramas

Intel Labs, the research division of semiconductor manufacturer Intel, has joined forces with Blockade Labs to collaborate on the development of LDM3D, a diffusion neural network capable of generating 360-degree panoramic 3D images based on textual input. The training of LDM3D involved utilising over 400 million pairs of images and their corresponding descriptions.

Moreover, the researchers have created a specialised application called DepthFusion, which combines 2D photographs and depth maps to generate images with a spherical view. They employed the Dense Prediction Transformer (DPT) model to accurately determine the depth of each pixel.

Representatives from Intel Labs have stated that generative AI technology empowers individuals to save time and unleash their creativity. They mentioned that unlike most existing AI models that are restricted to generating 2D images, LDM3D has the capability to generate both an image and a depth map based on a provided text prompt. They further explained that this approach provides a higher level of accuracy in terms of relative depth for each pixel compared to conventional post-processing methods, allowing developers to quickly create scenes.

The researchers believe that their groundbreaking advancement has the potential to revolutionise application development in various domains, including online gaming, interior design, architectural solutions, and virtual museums.