Hi-Fi: Hierarchical Feature Integration for Skeleton Detection
share
In natural images, skeleton scales (thickness) may significantly vary among objects and object parts. Thus, robust skeleton detection requires more powerful multi-scale feature integration ability than other vision tasks. In this paper, we present a new convolutional neural network (CNN) architecture by introducing a novel hierarchical feature integration mechanism, named Hi-Fi, to address the object skeleton detection problem. CNNs can capture high-level semantics from deeper layers as well as low-level details from shallower layers. By hierarchically integrating convolutional features with bi-direction guidance, our approach (1) enables mutual refinement across features of different levels, and (2) possesses strong ability to capture both rich object context and high-resolution details. Experimental results on several benchmarks show that our method significantly outperforms the state-of-the-art methods in terms of effectively fusing features from very different scales, as evidenced by a considerable performance improvement on skeleton detection. Our method also refreshes the state-of-the-art performance of boundary detection task.
READ FULL TEXT