DE-TensoRF: Data-Efficient and Fast NeRFs
April, 2023


Developed DE-TensoRF, a model that can render high-quality 3D objects with as few as 3 input images, and in under 15 min on a single GPU. Proposed three techniques to achieve data-effiency: (i) Obtain unseen views using symmetry (ii) Semantic conditioning (iii) Semantic Loss.
GitHub | Report |