Embedded Files
Abstract
Abstract
Video object segmentation aims to segment objects in a video sequence, given some user annotation which indicates the object of interest. Although Convolutional Neural Networks (CNNs) have been used in the recent past for the purpose of foreground segmentation in videos, adversarial training methods have not been used effectively to solve this problem, in spite of its extensive use for solving many other problems in Computer Vision. Earlier, flow features and motion trajectories have been extensively used to capture the temporal consistency between subsequent frames to segment moving objects in videos. However, we show that our proposed framework of processing the video frames independently using a deep generative adversarial network (GAN), is able to maintain the temporal coherency across frames without the use of any explicit trajectory based information, to provide superior results. Our main contribution lies in introducing a GAN based framework along with the incorporation of an Intersection-over-Union score based novel cost function for training the model, to solve the problem of foreground object segmentation in videos. The proposed method, when evaluated on popular real-world video segmentation datasets viz. DAVIS, SegTrack-v2 and YouTube-Objects, exhibits substantial performance gain over the recent state-of-the-art methods.
Video object segmentation aims to segment objects in a video sequence, given some user annotation which indicates the object of interest. Although Convolutional Neural Networks (CNNs) have been used in the recent past for the purpose of foreground segmentation in videos, adversarial training methods have not been used effectively to solve this problem, in spite of its extensive use for solving many other problems in Computer Vision. Earlier, flow features and motion trajectories have been extensively used to capture the temporal consistency between subsequent frames to segment moving objects in videos. However, we show that our proposed framework of processing the video frames independently using a deep generative adversarial network (GAN), is able to maintain the temporal coherency across frames without the use of any explicit trajectory based information, to provide superior results. Our main contribution lies in introducing a GAN based framework along with the incorporation of an Intersection-over-Union score based novel cost function for training the model, to solve the problem of foreground object segmentation in videos. The proposed method, when evaluated on popular real-world video segmentation datasets viz. DAVIS, SegTrack-v2 and YouTube-Objects, exhibits substantial performance gain over the recent state-of-the-art methods.
Paper
Paper
VidSeg-GAN: Generative Adversarial Network for Video Object Segmentation Tasks
VidSeg-GAN: Generative Adversarial Network for Video Object Segmentation Tasks
In 11th Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP), Hyderabad, India, 2018.
In 11th Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP), Hyderabad, India, 2018.
Architecture
Architecture
Fig.1: Proposed VidSeg-GAN architecture for Video Object Segmentation. GT refers to GroundTruth).
Quantitative Results
Quantitative Results
Qualitative Results
Qualitative Results
Video result
Video result
RGB
MSK (CVPR. 2017)
VidSeg-GAN (Ours)
(ICVGIP, 2018)
Ground Truth
Page updated
Google Sites
Report abuse