RecConv: Efficient Recursive Convolutions for Multi-Frequency Representations

Mingshu Zhao, Yi Luo, Yong Ouyang · 2024

Read RecConv: Efficient Recursive Convolutions for Multi-Frequency Representations by Mingshu Zhao, Yi Luo, Yong Ouyang (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-2412-19628.txt.

Description

Mingshu Zhao, Yi Luo, and Yong Ouyang build large image filters from stacked small ones with a recursive split, in a paper from two thousand twenty-four. Large filters widen the view of a vision net but balloon in size and cost. Their method, RecConv, grows view by levels while cost stays flat per level.

Parameter growth stays linear in levels, far below the steep growth of standard large filters. Compute rises only slightly at the top level in their counts. A test net beat a close rival on object finding by near two points at like cost. Code and models are shared. Readers who design vision nets get wide views at small-filter prices.

Catalog record

CREATORMingshu Zhao, Yi Luo, Yong Ouyang
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-19628

Browse the full library · About the Tower of Babel