Computer Vision News - June 2016

Every month, Computer Vision News reviews a research from our field. This month we have chosen to review Pull the Plug? Predicting If Computers or Humans Should Segment Images , a research paper presenting a resource allocation framework for estimating how best to allocate human effort in order to help computer follow-up with higher quality algorithm-drawn segmentations. The proposed model offers a sharp time saving in the human effort without compromising on segmentation quality. The paper will be presented in just a few weeks at CVPR 2016 and you can read it here . We are indebted to the authors ( Danna Gurari, Suyog Dutt Jain, Margrit Betke and Kristen Grauman ) for generously providing never published before images to help illustrate this review. Purpose Automatic segmentation algorithms have already generated huge benefits in procedures which were previously performed by tedious, manual human work. Nowadays, image segmentation algorithms can produce high-quality foreground object segmentations when they are successful, yet their performance is often inconsistent when applied on diverse datasets. A common question asked by people who need to annotate images is whether automated options are sufficient for their images or some degree of human intervention is required to achieve accurate annotations. A group of researchers have investigated the problem: their method predicts for each image in a batch, whether to “Pull The Plug” (PTP) on humans or computers. In other words, the system predicts whether the annotation should come from human or computer. The authors of the paper believe that this is the first study offering a practical recommendation about when to “Pull The Plug” on humans or computers for image segmentation. Computer Vision News Pull the Plug Computer Vision News Research 17 Figure 1