Dense Recurrent Neural Networks for Scene Labeling

01/21/2018
by   Heng Fan, et al.
0

Recently recurrent neural networks (RNNs) have demonstrated the ability to improve scene labeling through capturing long-range dependencies among image units. In this paper, we propose dense RNNs for scene labeling by exploring various long-range semantic dependencies among image units. In comparison with existing RNN based approaches, our dense RNNs are able to capture richer contextual dependencies for each image unit via dense connections between each pair of image units, which significantly enhances their discriminative power. Besides, to select relevant and meanwhile restrain irrelevant dependencies for each unit from dense connections, we introduce an attention model into dense RNNs. The attention model enables automatically assigning more importance to helpful dependencies while less weight to unconcerned dependencies. Integrating with convolutional neural networks (CNNs), our method achieves state-of-the-art performances on the PASCAL Context, MIT ADE20K and SiftFlow benchmarks.

READ FULL TEXT

page 1

page 7

research
11/09/2018

Scene Parsing via Dense Recurrent Neural Networks with Attentional Selection

Recurrent neural networks (RNNs) have shown the ability to improve scene...
research
09/02/2015

DAG-Recurrent Neural Networks For Scene Labeling

In image labeling, local representations for image units are usually gen...
research
11/22/2016

Scene Labeling using Gated Recurrent Units with Explicit Long Range Conditioning

Recurrent neural network (RNN), as a powerful contextual dependency mode...
research
08/27/2018

Why Self-Attention? A Targeted Evaluation of Neural Machine Translation Architectures

Recently, non-recurrent architectures (convolutional, self-attentional) ...
research
06/16/2021

Invertible Attention

Attention has been proved to be an efficient mechanism to capture long-r...
research
07/08/2016

Multi-level Contextual RNNs with Attention Model for Scene Labeling

Context in image is crucial for scene labeling while existing methods on...
research
03/09/2017

DA-RNN: Semantic Mapping with Data Associated Recurrent Neural Networks

3D scene understanding is important for robots to interact with the 3D w...

Please sign up or login with your details

Forgot password? Click here to reset