Curiosity Guided Fine-Tuning for Encoder-Decoder-Based Visual Forecasting

Yuta KAMIKAWA; Atsushi HASHIMOTO; Motoharu SONOGASHIRA; Masaaki IIYAMA

doi:10.1587/transinf.2020EDP7166

Curiosity Guided Fine-Tuning for Encoder-Decoder-Based Visual Forecasting

Yuta KAMIKAWA, Atsushi HASHIMOTO, Motoharu SONOGASHIRA, Masaaki IIYAMA

Full Text Views

0

Cite this

Summary :

An encoder-decoder (Enc-Dec) model is one of the fundamental architectures in many computer vision applications. One desired property of a trained Enc-Dec model is to feasibly encode (and decode) diverse input patterns. Aiming to obtain such a model, in this paper, we propose a simple method called curiosity-guided fine-tuning (CurioFT), which puts more weight on uncommon input patterns without explicitly knowing their frequency. In an experiment, we evaluated CurioFT in a task of future frame generation with the CUHK Avenue dataset and found that it reduced the mean square error by 7.4% for anomalous scenes, 4.8% for common scenes, and 6.6% in total. Some other experiments with the UCSD dataset further supported the reasonability of the proposed method.

Publication: IEICE TRANSACTIONS on Information Vol.E104-D No.5 pp.752-761

Publication Date: 2021/05/01

Publicized: 2021/02/02

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2020EDP7166

Type of Manuscript: PAPER

Category: Image Recognition, Computer Vision

Authors

Yuta KAMIKAWA
  Kyoto University,OMRON SINIC X Corp.
Atsushi HASHIMOTO
  OMRON SINIC X Corp.
Motoharu SONOGASHIRA
  Kyoto University
Masaaki IIYAMA
  Kyoto University

Keyword

imbalance data, pixel-wise prediction, visual forecasting

Cite this

Copy

Yuta KAMIKAWA, Atsushi HASHIMOTO, Motoharu SONOGASHIRA, Masaaki IIYAMA, "Curiosity Guided Fine-Tuning for Encoder-Decoder-Based Visual Forecasting" in IEICE TRANSACTIONS on Information, vol. E104-D, no. 5, pp. 752-761, May 2021, doi: 10.1587/transinf.2020EDP7166.
Abstract: An encoder-decoder (Enc-Dec) model is one of the fundamental architectures in many computer vision applications. One desired property of a trained Enc-Dec model is to feasibly encode (and decode) diverse input patterns. Aiming to obtain such a model, in this paper, we propose a simple method called curiosity-guided fine-tuning (CurioFT), which puts more weight on uncommon input patterns without explicitly knowing their frequency. In an experiment, we evaluated CurioFT in a task of future frame generation with the CUHK Avenue dataset and found that it reduced the mean square error by 7.4% for anomalous scenes, 4.8% for common scenes, and 6.6% in total. Some other experiments with the UCSD dataset further supported the reasonability of the proposed method.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2020EDP7166/_p

Copy

@ARTICLE{e104-d_5_752,
author={Yuta KAMIKAWA, Atsushi HASHIMOTO, Motoharu SONOGASHIRA, Masaaki IIYAMA, },
journal={IEICE TRANSACTIONS on Information},
title={Curiosity Guided Fine-Tuning for Encoder-Decoder-Based Visual Forecasting},
year={2021},
volume={E104-D},
number={5},
pages={752-761},
abstract={An encoder-decoder (Enc-Dec) model is one of the fundamental architectures in many computer vision applications. One desired property of a trained Enc-Dec model is to feasibly encode (and decode) diverse input patterns. Aiming to obtain such a model, in this paper, we propose a simple method called curiosity-guided fine-tuning (CurioFT), which puts more weight on uncommon input patterns without explicitly knowing their frequency. In an experiment, we evaluated CurioFT in a task of future frame generation with the CUHK Avenue dataset and found that it reduced the mean square error by 7.4% for anomalous scenes, 4.8% for common scenes, and 6.6% in total. Some other experiments with the UCSD dataset further supported the reasonability of the proposed method.},
keywords={},
doi={10.1587/transinf.2020EDP7166},
ISSN={1745-1361},
month={May},}

Copy

TY - JOUR
TI - Curiosity Guided Fine-Tuning for Encoder-Decoder-Based Visual Forecasting
T2 - IEICE TRANSACTIONS on Information
SP - 752
EP - 761
AU - Yuta KAMIKAWA
AU - Atsushi HASHIMOTO
AU - Motoharu SONOGASHIRA
AU - Masaaki IIYAMA
PY - 2021
DO - 10.1587/transinf.2020EDP7166
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 5
JA - IEICE TRANSACTIONS on Information
Y1 - May 2021
AB - An encoder-decoder (Enc-Dec) model is one of the fundamental architectures in many computer vision applications. One desired property of a trained Enc-Dec model is to feasibly encode (and decode) diverse input patterns. Aiming to obtain such a model, in this paper, we propose a simple method called curiosity-guided fine-tuning (CurioFT), which puts more weight on uncommon input patterns without explicitly knowing their frequency. In an experiment, we evaluated CurioFT in a task of future frame generation with the CUHK Avenue dataset and found that it reduced the mean square error by 7.4% for anomalous scenes, 4.8% for common scenes, and 6.6% in total. Some other experiments with the UCSD dataset further supported the reasonability of the proposed method.
ER -