You may have seen relatable AI before, but this one will give more than facial reconstruction. This is the chance for all to explore this AI experiment converting 2D images to 3D. If you try to train an AI to convert any type of image poses a challenge then Fleisher said:
It is varied in a few ways and same in others. While the mechanics of constructing and to train a Neural Network can poses similarities, the mission of creating a model which will understand human body and environments require a lot of data. In the background, a few models are being trained for a few other tasks and developing a classification method to identify the dominant elements of the image. Once we know what we can expect we can pick a model more refined for how it looks in 3D.
You can get the chance to explore this AI experiment converting 2D images to 3D. On this issue, Fleisher also says that:
It is necessary to say that we do not share any data users upload. However, use it for evaluating the machine learning models for getting a better understanding of what works. We also encourage the users for sharing their Volume reconstructions socially using the share icon from the top.
They are developing Volume as an end-to-end solution to convert flat 2D images into 3D images. Though the API is yet in early development, it is easily seen as a go-to application for video editors. Quite soon, game developers will want to use it as well. In the words of Fleisher, Volume will continue getting newer features:
We are looking into more popular areas like image augmentation. We did experiments and built an application called ReTouch. This enables the use of depth estimation for retouching images in 2.5D. By building Volume as a cloud, computer vision API we are able to provide the same results we use for 3D reconstruction.
You get to check the Volume API and convert your images into 3D using the team’s website.
