Faster gaze prediction with dense networks and Fisher pruning (2018-01-17T00:00:00.000000Z)

TL;DR

Through a combination of knowledge distillation and Fisher pruning, this paper obtains much more runtime-efficient architectures for saliency prediction, achieving a 10x speedup for the same AUC performance as a state of the art network on the CAT2000 dataset.

Abstract

Predicting human fixations from images has recently seen large improvements by leveraging deep representations which were pretrained for object recognition. However, as we show in this paper, these networks are highly overparameterized for the task of fixation prediction. We first present a simple yet principled greedy pruning method which we call Fisher pruning. Through a combination of knowledge distillation and Fisher pruning, we obtain much more runtime-efficient architectures for saliency prediction, achieving a 10x speedup for the same AUC performance as a state of the art network on the CAT2000 dataset. Speeding up single-image gaze prediction is important for many real-world applications, but it is also a crucial step in the development of video saliency models, where the amount of data to be processed is substantially larger.

Authors

Alykhan Tejani

3 papers

Ferenc Huszár

4 papers

Lucas Theis

5 papers

TL;DR

Abstract

Authors

References32 items

Saliency Benchmarking: Separating Models, Maps and Metrics

Predicting Human Eye Fixations via an LSTM-Based Saliency Attentive Model

Pruning Convolutional Neural Networks for Resource Efficient Inference

Understanding deep learning requires rethinking generalization

Exploiting inter-image similarity and ensemble of extreme learners for fixation prediction using deep features

A Deep Spatial Contextual Long-Term Recurrent Convolutional Network for Saliency Detection

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

Pruning Filters for Efficient ConvNets

Densely Connected Convolutional Networks

Spatio-Temporal Modeling and Prediction of Visual Attention in Graphical User Interfaces

What Do Different Evaluation Metrics Tell Us About Saliency Models?

Shallow and Deep Convolutional Networks for Saliency Prediction

SALICON: Reducing the Semantic Gap in Saliency Prediction by Adapting Deep Neural Networks

DeepFix: A Fully Convolutional Neural Network for Predicting Human Eye Fixations

Learning both Weights and Connections for Efficient Neural Network

SALICON: Saliency in Context

CAT2000: A Large Scale Fixation Dataset for Boosting Saliency Research

Distilling the Knowledge in a Neural Network

Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification

Adam: A Method for Stochastic Optimization

Deep Gaze I: Boosting Saliency Prediction with Feature Maps Trained on ImageNet

Very Deep Convolutional Networks for Large-Scale Image Recognition

Large-Scale Optimization of Hierarchical Features for Saliency Prediction in Natural Images

Saliency Based Image Cropping

Gaze-Driven video streaming with saliency-based dual-stream switching

A Benchmark of Computational Models of Saliency to Predict Human Fixations

Rectified Linear Units Improve Restricted Boltzmann Machines

Learning to predict where humans look

Second Order Derivatives for Network Pruning: Optimal Brain Surgeon

Gradient-based learning applied to document recognition

Optimal Brain Damage

Shifts in selective visual attention: towards the underlying neural circuitry.

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names