Experiments on theĬaltech Pedestrians dataset show that our approach reduces the number of (Q-net) that sequentially selects regions to zoom in. Built upon reinforcement learning, our approach consists ofĪ model (R-net) that uses coarse detection results to predict the potentialĪccuracy gain for analyzing a region at a higher resolution and another model Detection progresses in aĬoarse-to-fine manner, first on a down-sampled version of the image and then onĪ sequence of higher resolution regions identified as likely to improve theĭetection accuracy. Varied sizes appear in high resolution images.
Object detection while retaining accuracy for scenarios where objects with Davis Download PDF Abstract: We introduce a generic framework that reduces the computational cost of
Authors: Mingfei Gao, Ruichi Yu, Ang Li, Vlad I.
0 kommentar(er)
