INRet: A General Framework for Accurate Retrieval of INRs for Shapes

Gavin Guan, Yuchen Chen, Haoze Deng, Daniel Kwan, Ruofan Liang, Selvakumar Panneer, Nilesh Jain, Nandita Vijaykumar

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: INRet: A General Framework for Accurate Retrieval of INRs for Shapes

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: A framework for retrieval of shapes represented with implicit neural representations (INRs), supporting a diverse set of INR architectures and implicit functions

Abstract: Implicit neural representations (INRs) have become an important representation for encoding various data types, such as 3D objects/scenes, videos, and audio. They have proven to be particularly effective at generating 3D content, e.g., 3D scene reconstruction from 2D images, novel content creation, as well as the representation, interpolation and completion of 3D shapes. With the widespread generation of 3D data in an INR format, there is a need to support effective organization and retrieval of INRs saved in a data store. A key aspect of retrieval and clustering of INRs in a data store is the formulation of similarity between INRs that would, for example, enable retrieval of similar INRs using a query INR. In this work, we propose INRet (INR Retrieve), a method for determining similarity between INRs that represent shapes, thus enabling accurate retrieval of similar shape INRs from an INR data store. INRet flexibly supports different INR architectures such as INRs with octree grids and hash grids, as well as different implicit functions including signed/unsigned distance function and occupancy field. We demonstrate that our method is more general and accurate than the existing INR retrieval method, which only supports simple MLP INRs and requires the same architecture between the query and stored INRs. Compared to 3D shape retrieval by converting INRs to other representations like point clouds or multi-view images, INRet achieves higher retrieval accuracy while avoiding the overhead of conversion.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7597