Deep Learning Techniques to Easily Convert 2-D Sketches into 3-D Shapes

Deep Learning Techniques to Easily Convert 2-D Sketches into 3-D Shapes


A team of researchers from the University of Hong Kong, University of British Columbia, and Microsoft designed a novel technique to convert 2-D into 3-D sketches easily. The researchers further revealed that our deep learning method could enable artists to outline a base shape quickly. This base shape can be further fine-tuned with automated image vectorization software. In addition to this, new users may also access this software to from 3-D content and images.

What are the Different Techniques Used to form 3D Sketches?

This novel technique is based on geometric – based 3-D modeling sketch work and works as per defined sketch inputs. The team also revealed that they want to design a method which is both generic and intelligent. By generic, it means that we don’t have to train a new machine learning statistical model for specific categories. By intelligent, it means that the software should require fewer inputs and should automatically provide inference. Researchers in a statement mentioned that it is not easy to concert an imperfect and poor 2-D image to an advanced 3-D shape.

One of the primary challenge faced by them is the disparate information existing between the sketch stages and 3-D model. These stages include features such as bends, bumps, ridge surfaces, valleys, and creases. All these features can be filled in by using a convolutional neural network (CNN) technique to anticipate the 3-D surface area. The CNN technique developed by researchers is trained on a huge dataset produced by portraying several 3-D sketches using NPR or Non-photorealistic line Rendering. The NPR technique automatically imitates different drawings of freeform shapes. The researchers also validated this approach a robust and user-friendly as compared to 2-D sketches.

Future Improvisation of Tool to Make Sketching Easy For Artists

The team of researchers opined that going forward we would like to explore various other possibilities of their deep learning algorithms. They also hope to improve the software to handle numerous scale modeling. Thus, making it easier for artists to sketch large shapes with perfection and ease.