Revisiting Video Saliency: A Large-scale Benchmark and a New Model

Abstract

In this work, we contribute to video saliency research in two ways. First, we introduce a new benchmark for predicting human eye movements during dynamic scene free-viewing, which is long-time urged in this field. Our dataset, named DHF1K (Dynamic Human Fixation), consists of 1K high-quality, elaborately selected video sequences spanning a large range of scenes, motions, object types and background complexity. Existing video saliency datasets lack variety and generality of common dynamic scenes and fall short in covering challenging situations in unconstrained environments. In contrast, DHF1K makes a significant leap in terms of scalability, diversity and difficulty, and is expected to boost video saliency modeling. Second, we propose a novel video saliency model that augments the CNN-LSTM network architecture with an attention mechanism to enable fast, end-to-end saliency learning. The attention mechanism explicitly encodes static saliency information, thus allowing LSTM to focus on learning more flexible temporal saliency representation across successive frames. Such a design fully leverages existing large-scale static fixation datasets, avoids overfitting, and significantly improves training efficiency and testing performance. We thoroughly examine the performance of our model, with respect to state-of-the-art saliency models, on three large-scale datasets (i.e., DHF1K, Hollywood2, UCF sports). Experimental results over more than 1.2K testing videos containing 400K frames demonstrate that our model outperforms other competitors.

arXiv commentCVPR2018 paper. Website: https://github.com/wenguanwang/DHF1K We have corrected some statistics of our results (baseline training setting (iii)) on UCF sports dataset

References in corpus (10)

• Adam: A Method for Stochastic Optimization
• Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting
• Deep Residual Learning for Image Recognition
• LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop
• Video Salient Object Detection via Fully Convolutional Networks
• Stacked Attention Networks for Image Question Answering
• What is a salient object? A dataset and a baseline model for salient object detection
• Shallow and Deep Convolutional Networks for Saliency Prediction
• Image Question Answering using Convolutional Neural Network with Dynamic Parameter Prediction
• Deep Cropping via Attention Box Prediction and Aesthetics Assessment

Cited by in corpus (12)

• DAVE: A Deep Audio-Visual Embedding for Dynamic Saliency Prediction
• Improving Video Compression With Deep Visual-Attention Models
• ATSal: An Attention Based Architecture for Saliency Prediction in 360 Videos
• Attention-SLAM: A Visual Monocular SLAM Learning from Human Gaze
• Predicting video saliency using crowdsourced mouse-tracking data
• Global and Local Sensitivity Guided Key Salient Object Re-augmentation for Video Saliency Detection
• Temporal Saliency Adaptation in Egocentric Videos
• SUSiNet: See, Understand and Summarize it
• ViP: Video Platform for PyTorch
• Model-guided Multi-path Knowledge Aggregation for Aerial Saliency Prediction
• Supersaliency: A Novel Pipeline for Predicting Smooth Pursuit-Based Attention Improves Generalizability of Video Saliency
• CrowdFix: An Eyetracking Dataset of Real Life Crowd Videos