| Meno: | Filip
|
|---|
| Priezvisko: | Jurčák
|
|---|
| Názov: | Material Picker: Material recognition in images using deep learning
|
|---|
| Vedúci: | prof. RNDr. Roman Ďurikovič, PhD
|
|---|
| Rok: | 2020
|
|---|
| Kµúčové slová: | Deep learning, inverse rendering, material segmentation, machine learning, computer graphics, material recognition
|
|---|
| Abstrakt: | The process of setting material properties for realistic appearance is usually tiresome and often demands skill for fine-tuning the parameters, as different combinations of these parameters can produce different materials. To simplify this process, we introduce a pipeline consisting of deep neural networks to segment material and predict intrinsic scene characteristics, like diffuse and specular albedo, surface normals, glossiness, view vector, and illumination from a single image. Our pipeline thus provides solution to two of the most fundamental problems in computer vision and computer graphics - inverse rendering and material segmentation. We trained the networks on the dataset generated using physically-based techniques to ensure good generalization on real images.
|
|---|