--- permalink: projects/SVP --- S-Lab

Surveillance Video Parsing with Single Frame Supervision

Paper

  1. Si Liu , Changhu Wang, Ruihe Qian, Han Yu, Renda Bao, Surveillance Video Parsing with Single Frame Supervision, arxiv preprint[ Demo].[ Paper]

Dataset

You can download the Indoor Dataset from Baidu Drive(pwd: 7mbh).

Introduction

  Surveillance video parsing, which segments the video frames into several labels, e.g., face, pants, left-leg, has wide applications. However, pixel-wisely annotating all frames is tedious and inefficient. In this paper, we develop a Single frame Video Parsing (SVP) method which requires only one labeled frame per video in training stage. To parse one particular frame, the video segment preceding the frame is jointly considered. SVP (i) roughly parses the frames within the video segment, (ii) estimates the optical flow between frames and (iii) fuses the rough parsing results warped by optical flow to produce the refined parsing result. The three components of SVP, namely frame parsing, optical flow estimation and temporal fusion are integrated in an end-to-end manner. Experimental results on two surveillance video datasets

Deep Architecture

  Figure 1 shows the SVP network.The input is a triplet {I(t-l),I(t-s),I(t)}, among which only It is labeled. l and s are set empirically. The output is the parsing result P(t). SVP contains three sub-networks. As a preprocessing step, we use Faster R-CNN to extract the human region. Then, the triplet are fed into Conv1~Conv5 for discriminative feature extraction. The frame parsing sub-network produces the rough labelmaps for the triplet, denoted as{P(t-l),P(t-s),P(t)}. The optical flow estimation sub-network aims to estimate the dense correspondence between adjacent frames.

Experiments

References

  • [1] V. Badrinarayanan, A. Handa, and R. Cipolla. Segnet: A deep convolutional encoder-decoder architecture for robust semantic pixel-wise labelling. arXiv:1505.07293, 2015.
  • [2] M. Bai, W. Luo, K. Kundu, and R. Urtasun. Exploiting semantic information and deep matching for optical flow. In ECCV, 2016.
  • [3] A. Bearman, O. Russakovsky, V. Ferrari, and L. Fei-Fei. What’s the point: Semantic segmentation
  • ......