2019/08/13 by Jungbeom Lee, Eunji Kim, Lee, Jungbeom +7
Computer Science · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Multimodal Machine Learning Applications
paper · pdf · doi:10.48550/arxiv.1908.04501
openalex publication_date 2019/08/13 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
When a deep neural network is trained on data with only image-level labeling,\nthe regions activated in each image tend to identify only a small region of the\ntarget object. We propose a method of using videos automatically harvested from\nthe web to identify a larger region of the target object by using temporal\ninformation, which is not present in the static image. The temporal variations\nin a video allow different regions of the target object to be activated. We\nobtain an activated region in each frame of a video, and then aggregate the\nregions from successive frames into a single image, using a warping technique\nbased on optical flow. The resulting localization maps cover more of the target\nobject, and can then be used as proxy ground-truth to train a segmentation\nnetwork. This simple approach outperforms existing methods under the same level\nof supervision, and even approaches relying on extra annotations. Based on\nVGG-16 and ResNet 101 backbones, our method achieves the mIoU of 65.0 and 67.4,\nrespectively, on PASCAL VOC 2012 test images, which represents a new\nstate-of-the-art.\n