ProtoQuant: Quantization of Prototypical Parts For General and Fine-Grained Image Classification

Mikołaj Janusz, Adam Wróbel, Bartosz Zieliński, Dawid Rymarczyk · 2026

Read ProtoQuant: Quantization of Prototypical Parts For General and Fine-Grained Image Classification by Mikołaj Janusz, Adam Wróbel, Bartosz Zieliński, Dawid Rymarczyk (2026) 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-2602-06592.txt.

Description

ProtoQuant, a neural network design from Mikołaj Janusz and three co-authors, classifies images by matching parts of a picture to learned example parts. Models of this kind are valued because their choices can be explained in plain terms: a part is classed by what it looks like. Earlier versions struggled on very large image sets and needed costly retraining.

ProtoQuant keeps its example parts fixed inside a learned codebook, which stops them from drifting away from real training images over time. It also works without retraining the main body of the network, so it acts as a light, clear add-on head. Tests on ImageNet and on fine-grained sets of birds and cars show accuracy close to rival methods, with explanations that stay grounded in real data.

Catalog record

CREATORMikołaj Janusz, Adam Wróbel, Bartosz Zieliński, Dawid Rymarczyk
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-06592

Browse the full library · About the Tower of Babel