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

By Wenguan Wang, Jianbing Shen, Fang Guo, Ming-Ming Cheng, Ali Borji. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

305 citations recorded by the source corpus.

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.

Subfields: Computer Vision and Pattern Recognition, Human-Computer Interaction

View Paper: https://www.semanticscholar.org/paper/fdf8c9c4c30c6005c2f0e92ce9db3de5ab8b5d29

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Human-Computer Interaction

Citations in the corpus

Human Gaze Assisted Artificial Intelligence: A Review (2020, background): “…became available for images [Papadopoulos et al. , 2014; Li et al. , 2014; Xu et al. , 2014; Bylinskii et al. , 2015b; Bylinskii et al. , 2015a; Krafka et al. , 2016], videos [Mathe and Sminchisescu, 2014; Wang et al. , 2018 ] , and 360-degree videos [ Zhang et al. , 2018b; Xu et al. , 2018].”
Revisiting Video Saliency Prediction in the Deep Learning Era (2021, background): “A preliminary version of this work has appeared in CVPR 2018 [1].”
Spatiotemporal Knowledge Distillation for Efficient Estimation of Aerial Video Saliency (2019, methodology): “iSEEL [64], DVA [10], SalNet [50], STS [57] and ACLNet [61].”
Pyramid Dilated Deeper ConvLSTM for Video Salient Object Detection (2018, methodology): “, eye fixation prediction [41, 39] and salient object detection [49, 47].”
Learning Unsupervised Video Object Segmentation Through Visual Attention (2019, methodology): “Recently, deep learning based attention models [71, 28, 55, 73, 75] were proposed and generally yield better performance.”
Towards Large-Scale Small Object Detection: Survey and Benchmarks (2022, methodology): “As stated in [146], [147], the annotation quality of a dataset is the dominating factor for its lifespan as a benchmark, hence we install a two-step quality assurance procedure after the initial annotation.”
A Deep Network Solution for Attention and Aesthetics Aware Photo Cropping (2019, background): “Deep learning-based attention models [29], [30], [31], [32], [33], [34] become increasingly popular in recent years, driven by the success of deep learning in object recognition and large-scale visual attention dataset (e.”
Concealed Object Detection (2021, background): “As discussed in [11], [17], [51], the quality of annotation and size of a dataset are determining factors for its lifespan as a benchmark.”
Shifting More Attention to Video Salient Object Detection (2019, methodology): “The stimuli of DAVSOD come from DHF1K [73], which is the current largest-scale dynamic eye-tracking dataset.”