ID-to-3D: Expressive ID-guided 3D Heads via Score Distillation Sampling

Authors: Francesca Babiloni, Alexandros Lattas, Jiankang Deng, Stefanos Zafeiriou

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We assess the efficacy of ID-to-3D as a specialized method for ID-driven expressive human face generation in different scenarios and report comparative analysis against state-of-the-art text-to-3D and image-to-3D generation pipelines. Further analysis, implementation details and discussion can be found in additional material. Quantitative Comparisons: FID and User-Study. The evaluation of the generated 3D geometries and textures is performed using the Frechet Inception Distance (FID) metric [28].
Researcher Affiliation Academia Francesca Babiloni, Alexandros Lattas, Jiankang Deng, Stefanos Zafeiriou Imperial College London, UK
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes We base our method on public code, report our technical details in the supplemental materials, and will be releasing our code. We use already publicly available data, and will be open-sourcing our code.
Open Datasets Yes To overcome the need for a large-scale dataset, we leverage a small dataset of human heads with different expressions (NHPM) [24]
Dataset Splits Yes We include all the above details in the supplemental materials.
Hardware Specification Yes We have included the above details in the supplemental materials.
Software Dependencies Yes We have included the above details in the supplemental materials.
Experiment Setup Yes We include all the above details in the supplemental materials.