Technical Report for the 5th CLVision Challenge at CVPR: Addressing the Class-Incremental with Repetition using Unlabeled Data - 4th Place Solution

Panagiota Moraiti Efstathios Karypidis · 2025

Read Technical Report for the 5th CLVision Challenge at CVPR: Addressing the Class-Incremental with Repetition using Unlabeled Data - 4th Place Solution by Panagiota Moraiti Efstathios Karypidis (2025) 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-2503-15697.txt.

Description

Panagiota Moraiti and Efstathios Karypidis report their fourth place entry to a computer vision challenge on learning that never stops, in this 2025 technical report. The challenge setting lets classes reappear across rounds, and each round covers only part of the class list. Unlabeled images arrive too, and some belong to classes the model should ignore.

Their approach fights forgetting, the classic failure where new learning erases the old. It keeps a frozen copy of past knowledge through distillation and lets the model label unlabeled images for itself where it is confident. Accuracy averaged about seventeen percent in the selection phase and twenty-one percent in the final round, more than double the baseline. The authors share their code openly for others to build on.

Catalog record

CREATORPanagiota Moraiti Efstathios Karypidis
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE18 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-15697

Browse the full library · About the Tower of Babel