Visual Saliency Prediction Using a Mixture of Deep Neural Networks

Lina Karam; Samuel Dodge

RECORD

Visual Saliency Prediction Using a Mixture of Deep Neural Networks

Samuel Dodge, Lina Karam

Computer Vision and Pattern Recognition · Submitted 1 Feb 2017

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim We show that our mixture formulation leads to improvement in performance over an otherwise identical non-mixture model that does not incorporate global scene information.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 6,653 input tokens.

What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.08 · human participants 0.07 · controlled comparison 0.89 · randomized assignment 0.01 · states quantitative result 0.05 · compares with prior results 0.91 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; show every role.

Visual saliency models have recently begun to incorporate deep learning to achieve predictive capacity much greater than previous unsupervised methods. However, most existing models predict saliency using local mechanisms limited to the receptive field of the network. We propose a model that incorporates global scene semantic information in addition to local information gathered by a convolutional neural network. Our model is formulated as a mixture of experts. Each expert network is trained to predict saliency for a set of closely related images. The final saliency map is computed as a weighted mixture of the expert networks' output, with weights determined by a separate gating network. This gating network is guided by global scene information to predict weights. The expert networks and the gating network are trained simultaneously in an end-to-end manner. We show that our mixture formulation leads to improvement in performance over an otherwise identical non-mixture model that does not incorporate global scene information.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	Visual saliency models have recently begun to incorporate deep learning to achieve predictive capacity much greater than previous unsupervised methods.	background 1.00	0.04	0.03	asserted 1.00
s1	However, most existing models predict saliency using local mechanisms limited to the receptive field of the network.	gap 0.76	0.04	0.02	asserted 1.00
s2	We propose a model that incorporates global scene semantic information in addition to local information gathered by a convolutional neural network.	contribution 0.97	0.82	0.09	asserted 0.98
s3	Our model is formulated as a mixture of experts.	method 0.86	0.82	0.04	asserted 1.00
s4	Each expert network is trained to predict saliency for a set of closely related images.	method 0.99	0.83	0.03	asserted 1.00
s5	The final saliency map is computed as a weighted mixture of the expert networks' output, with weights determined by a separate gating network.	method 1.00	0.79	0.07	asserted 1.00
s6	This gating network is guided by global scene information to predict weights.	method 0.99	0.80	0.06	asserted 1.00
s7	The expert networks and the gating network are trained simultaneously in an end-to-end manner.	method 1.00	0.83	0.05	asserted 1.00
s8	We show that our mixture formulation leads to improvement in performance over an otherwise identical non-mixture model that does not incorporate global scene information.	result 1.00	0.96	0.85	asserted 1.00

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.83.

What was examined, and what was not

sources read

arXiv listing abstract v1

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

Visual saliency models have recently begun to incorporate deep learning to achieve predictive capacity much greater than previous unsupervised methods. However, most existing models predict saliency using local mechanisms limited to the receptive field of the network. We propose a model that incorporates global scene semantic information in addition to local information gathered by a convolutional neural network. Our model is formulated as a mixture of experts. Each expert network is trained to predict saliency for a set of closely related images. The final saliency map is computed as a weighted mixture of the expert networks' output, with weights determined by a separate gating network. This gating network is guided by global scene information to predict weights. The expert networks and the gating network are trained simultaneously in an end-to-end manner. We show that our mixture formulation leads to improvement in performance over an otherwise identical non-mixture model that does not incorporate global scene information.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

1 arXiv paper cite this in 2 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

What it cites · 5 sentences without a confident label

Builds on 1 paper

We allow our network to learn the center bias as in previous works [ref] , but also allow the center bias to vary between categories.

Deep Gaze I: Boosting Saliency Prediction with Feature Maps Trained on ImageNet · Proposed Approach · 0.89

Uses the method or tool of 1 paper

We use the Adadelta algorithm for optimizing the parameters of the neural network [ref] .

ADADELTA: An Adaptive Learning Rate Method · Training · 1.00

Uses data or values from 1 paper

We primarily use the CAT2000 dataset [ref] because it comprises a sufficient number of images for deep learning (2000 in total), and it sorts the images into 20 different categories.

CAT2000: A Large Scale Fixation Dataset for Boosting Saliency Research · Performance Evaluation · 0.98

Compares with 1 paper

Our model achieves the second best performance behind the DeepFix model [ref] .

DeepFix: A Fully Convolutional Neural Network for predicting Human Eye Fixations · Performance Evaluation · 1.00

Cites as background 3 papers

Deepfix [ref] improves performance by adding inception modules [ref] , using dilated convolutions, and explicitly modeling the center bias.

DeepFix: A Fully Convolutional Neural Network for predicting Human Eye Fixations · Introduction · 0.83
DeepGaze [ref] show that the deep features from VGG networks can be used without modification to predict saliency.

Deep Gaze I: Boosting Saliency Prediction with Feature Maps Trained on ImageNet · Introduction · 0.86
DeepGaze [ref] show that the deep features from VGG networks can be used without modification to predict saliency.

DeepGaze II: Reading fixations from deep features trained on object recognition · Introduction · 0.86

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.