Evaluating end-to-end and multi-stage 3D model generation with 3D morphable models: a comparative study on eyeglasses frames
| Year |
|---|
2023 |
| URI | Access Rights |
|---|---|
| https://hdl.handle.net/20.500.14911/210519 | |
| https://doi.org/10.15388/DAMSS.14.2023 | Viso teksto dokumentas (atviroji prieiga) / Full Text Document (Open Access) |
In the domain of product visualization, the ability to generate precise 3D models from 2D images is pivotal to enhancing user experience and engagement. Our investigation conducts an exhaustive comparative analysis of two distinct methodologies in generating 3D models of eyeglasses frames from a pair of product images: the end-to-end 3D model generation and a multi-stage approach utilizing 3D Morphable Models (3DMM). The end-to-end generation approach leverages deep learning algorithms to directly map input images into 3D models, ensuring an uninterrupted and expedited transformation. On the other hand, the multi-stage approach begins with the extraction of pertinent features from the 2D images, followed by segmentation and the application of 3DMM for shape reconstruction.