SeqViews2SeqLabels: Learning 3D Global Features via Aggregating Sequential Views by RNN with Attention

UM > Faculty of Science and Technology > DEPARTMENT OF COMPUTER AND INFORMATION SCIENCE

Status	已發表Published
	SeqViews2SeqLabels: Learning 3D Global Features via Aggregating Sequential Views by RNN with Attention
	Han, Z. Z.; Shang, M. Y.; Liu, Z. B.; Vong, C. M.; Liu, Y. S.; Han, J. W.; Chen, C. L. P.
	2019-02-01
Source Publication	IEEE Transactions on Image Processing (SCI-E)
ISSN	1057-7149
Pages	658-672
Abstract	It is a significant and popular research topic to learn 3D global features via aggregating multiple views for 3D shape analysis. In recent end-to-end training of deep learning models for 3D shapes, pooling is a widely adopted procedure of view aggregation. However, pooling is a procedure producing just max or mean value, which disregards the content information of almost all views and also the spatial information among views. To resolve these issues, Sequential Views To Sequential Labels (SeqViews2SeqLabels) is proposed as a novel deep learning model with an encoder-decoder structure based on Recurrent Neural Networks (RNNs) with attention. Within SeqViews2SeqLabels, its two connected parts, encoder-RNN followed by decoder-RNN, aim to learn the global features via aggregating sequential views and then perform shape classification upon the learned global features, respectively. Specifically, encoder-RNN learns the global features via simultaneously encoding the spatial and content information of sequential views, which captures the semantic meaning of view sequence.With newly proposed sequential labels, decoder-RNN performs more accurate classification upon the learned global features via predicting sequential labels step-bystep, which captures more discriminative information among shape classes. Moreover, the attention mechanism is introduced to further improve the discriminability of SeqViews2SeqLabels via weighting more on the low-level features of views which are distinctive to each shape class. Through the experimental results over three large-scale benchmarks, SeqViews2SeqLabels is verified to learn more discriminative global features via aggregating sequential view, which outperforms the state-of-theart methods for shape classification and retrieval.
Keyword	3D feature learning Sequential views Sequential labels View aggregation RNN Attention
Language	英語English
The Source to Article	PB_Publication
PUB ID	41190
Document Type	Journal article
Collection	DEPARTMENT OF COMPUTER AND INFORMATION SCIENCE
Corresponding Author	Liu, Y. S.
Recommended Citation GB/T 7714	Han, Z. Z.,Shang, M. Y.,Liu, Z. B.,et al. SeqViews2SeqLabels: Learning 3D Global Features via Aggregating Sequential Views by RNN with Attention[J]. IEEE Transactions on Image Processing (SCI-E), 2019, 658-672.
APA	Han, Z. Z.., Shang, M. Y.., Liu, Z. B.., Vong, C. M.., Liu, Y. S.., Han, J. W.., & Chen, C. L. P. (2019). SeqViews2SeqLabels: Learning 3D Global Features via Aggregating Sequential Views by RNN with Attention. IEEE Transactions on Image Processing (SCI-E), 658-672.
MLA	Han, Z. Z.,et al."SeqViews2SeqLabels: Learning 3D Global Features via Aggregating Sequential Views by RNN with Attention".IEEE Transactions on Image Processing (SCI-E) (2019):658-672.

Files in This Item:
There are no files associated with this item.

If you have any objections to this item, please fill out the form below and the administrator will contact you as soon as possible.
Content:
Email：	*
Affiliation No.
Verification Code:	Refresh

Any comments and suggestions are welcomed.
Title:	*
Content:
Email：	*
Verification Code:	Refresh