Cross Domain Adaptation using Adversarial networks with Cyclic loss

Manpreet Kaur, Ankur Tomar, Srijan Mishra, Shashwat Verma · 2024

Read Cross Domain Adaptation using Adversarial networks with Cyclic loss by Manpreet Kaur, Ankur Tomar, Srijan Mishra, Shashwat Verma (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-2412-01935.txt.

Description

Manpreet Kaur, Ankur Tomar, Srijan Mishra, and Shashwat Verma work on teaching image translators to carry their skill from one data domain to another, in this 2024 arXiv paper. Deep networks are brittle: small shifts between training and target data can wreck accuracy. The team studies generator networks that translate images across domains under an adversarial setup.

They experiment with activation choices and encoder-decoder layouts, and they add a new training signal, a cyclic loss, that keeps the translator honest on round trips. The aim is a translator that stays effective when paired with the original domain's network. Motivating uses include making labeled data from synthetic inputs without hand annotation, and widening where trained models can be trusted.

Catalog record

CREATORManpreet Kaur, Ankur Tomar, Srijan Mishra, Shashwat Verma
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-01935

Browse the full library · About the Tower of Babel