Nerfbusters: Removing Ghostly Artifacts from Casually Captured NeRFs
A method to clean up Neural Radiance Fields (NeRFs) for more accurate novel-view synthesis
Provides a method for cleaning up NeRFs to improve image quality for novel-view synthesis
Reference-based Image Composition with Sketch via Structure-aware Diffusion Model
A multi-input-conditioned image composition model that incorporates a sketch as a novel modal for image manipulation
Provides a method for image manipulation with the incorporation of a sketch
A Theory on Adam Instability in Large-Scale Machine Learning
A theory that explains divergent behavior noticed in the training of large language models and how it can be an artifact of the optimization algorithm used
Provides an explanation for divergent behavior in large language model training and how it can be addressed
Is ChatGPT a Good Recommender? A Preliminary Study
This paper explores the potential of ChatGPT to improve the performance of general-purpose recommendation models. Using prompts and few-shot prompting, the authors evaluate ChatGPT's performance on five recommendation scenarios and achieve promising results. Human evaluations show that ChatGPT can understand the provided information and generate clearer and more reasonable results.
ChatGPT has the potential to improve the performance of general-purpose recommendation models by leveraging extensive linguistic and world knowledge acquired from large-scale corpora. Few-shot prompting can inject interaction information that contains user potential interest and help ChatGPT better understand user needs and interests. Businesses can consider using ChatGPT for their recommendation systems to provide more accurate and relevant recommendations to their customers.
Tetra-NeRF: Representing Neural Radiance Fields Using Tetrahedra
This paper proposes using an adaptive representation based on tetrahedra and a Delaunay representation for Neural Radiance Fields (NeRFs) to improve novel view synthesis and 3D reconstruction. The approach combines concepts from 3D geometry processing, triangle-based rendering, and modern neural radiance fields and achieves better performance than voxel-based and point-based representations.
The use of tetrahedra and a Delaunay representation can improve the performance of Neural Radiance Fields for novel view synthesis and 3D reconstruction. Businesses that use 3D reconstruction or computer graphics can benefit from this approach to provide more detailed and accurate representations of scenes or objects.