End-to-end supermask pruning: Learning to prune image captioning models

Tan, Jia Huei and Chan, Chee Seng and Chuah, Joon Huang (2022) End-to-end supermask pruning: Learning to prune image captioning models. Pattern Recognition, 122. ISSN 0031-3203, DOI https://doi.org/10.1016/j.patcog.2021.108366.

Full text not available from this repository.


With the advancement of deep models, research work on image captioning has led to a remarkable gain in raw performance over the last decade, along with increasing model complexity and computational cost. However, surprisingly works on compression of deep networks for image captioning task has received little to no attention. For the first time in image captioning research, we provide an extensive comparison of various unstructured weight pruning methods on three different popular image captioning architectures, namely Soft-Attention, Up-Down and Object Relation Transformer . Following this, we propose a novel end-to-end weight pruning method that performs gradual sparsification based on weight sensitivity to the training loss. The pruning schemes are then extended with encoder pruning, where we show that conducting both decoder pruning and training simultaneously prior to the encoder pruning provides good overall performance. Empirically, we show that an 80% to 95% sparse network (up to 75% reduction in model size) can either match or outperform its dense counterpart. The code and pre-trained models for Up-Down and Object Relation Transformer that are capable of achieving CIDEr scores > 120 on the MSCOCO dataset but with only 8.7 MB and 14.5 MB in model size (size reduction of 96% and 94% respectively against dense versions) are publicly available at https://github.com/jiahuei/sparse- image-captioning . (c) 2021 Elsevier Ltd. All rights reserved.

Item Type: Article
Uncontrolled Keywords: Image captioning; Deep network compression; Deep learning
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
T Technology > TA Engineering (General). Civil engineering (General)
Divisions: Faculty of Engineering > Department of Chemical Engineering
Depositing User: Ms. Juhaida Abd Rahim
Date Deposited: 28 Apr 2022 00:27
Last Modified: 28 Apr 2022 00:27
URI: http://eprints.um.edu.my/id/eprint/33743

Actions (login required)

View Item View Item