INRFlow: Flow Matching for INRs in Ambient Space

Yuyang Wang, Anurag Ranjan, Josh Susskind, Miguel Angel Bautista · 2024

Read INRFlow: Flow Matching for INRs in Ambient Space by Yuyang Wang, Anurag Ranjan, Josh Susskind, Miguel Angel Bautista (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-03791.txt.

Description

Yuyang Wang, Anurag Ranjan, Josh Susskind, and Miguel Angel Bautista present a 2024 paper on generative models that work directly on raw data of many kinds. Flow matching models create images, videos, and even protein structures by learning a flowing transformation. Usually a compressor first squeezes data into a latent space, and each data type needs its own hand-built compressor. Their method, INRFlow, skips that stage entirely.

Borrowing from implicit neural representations, it trains a flow transformer with a pointwise objective in the original coordinate space, so predictions stay continuous. One architecture then handles images, 3D point clouds, and protein structures alike. Experiments show strong results in each domain, beating comparable two-stage approaches. The authors present INRFlow as a step toward one generative recipe for many data types, with no bespoke compression stage to design per field.

Catalog record

CREATORYuyang Wang, Anurag Ranjan, Josh Susskind, Miguel Angel Bautista
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-03791

Browse the full library · About the Tower of Babel