Self-Calibrating Photometric Stereo for Non-Lambertian Surfaces

Professor Wong, Kenneth Kwan Yee   (Principal Investigator (PI))

36

2019-09-01

2022-08-31

518999

Self-Calibrating Photometric Stereo for Non-Lambertian Surfaces

deep learning, non-Lambertian surfaces, photometric stereo, uncalibrated illuminations

VisionArtificial Intelligence and Machine learning

Engineering (E)

17203119

General Research Fund (GRF)

2019

Completed

1 Design and render a large scale synthetic dataset with realistic shapes and spatially-varying BRDFs for learning non-Lambertian photometric stereo under directional lighting. 2 Design and develop a neural network for estimating the surface normal of an object given photometric stereo images where light directions and intensities are known, and the network is robust to objects with spatially-varying materials. 3 Design and develop a neural network for estimating light directions and intensities given photometric stereo images only. 4 Design and develop a neural network for estimating the surface normal of an object given photometric stereo images with inaccurate light directions and intensities, and the network is robust to inaccurate lighting information. 5 Design and render a large scale synthetic dataset for photometric stereo under natural illumination. 6 Design and develop a neural network for estimating the surface normal of an object given photometric stereo images captured under natural illumination where environment lighting is known. 7 Design and develop a neural network for estimating the surface normal of an object given photometric stereo images captured under natural illumination where environment lighting is unknown