English

CARAFE++: Unified Content-Aware ReAssembly of FEatures

Computer Vision and Pattern Recognition 2020-12-10 v1

Abstract

Feature reassembly, i.e. feature downsampling and upsampling, is a key operation in a number of modern convolutional network architectures, e.g., residual networks and feature pyramids. Its design is critical for dense prediction tasks such as object detection and semantic/instance segmentation. In this work, we propose unified Content-Aware ReAssembly of FEatures (CARAFE++), a universal, lightweight and highly effective operator to fulfill this goal. CARAFE++ has several appealing properties: (1) Unlike conventional methods such as pooling and interpolation that only exploit sub-pixel neighborhood, CARAFE++ aggregates contextual information within a large receptive field. (2) Instead of using a fixed kernel for all samples (e.g. convolution and deconvolution), CARAFE++ generates adaptive kernels on-the-fly to enable instance-specific content-aware handling. (3) CARAFE++ introduces little computational overhead and can be readily integrated into modern network architectures. We conduct comprehensive evaluations on standard benchmarks in object detection, instance/semantic segmentation and image inpainting. CARAFE++ shows consistent and substantial gains across all the tasks (2.5% APbox, 2.1% APmask, 1.94% mIoU, 1.35 dB respectively) with negligible computational overhead. It shows great potential to serve as a strong building block for modern deep networks.

Keywords

Cite

@article{arxiv.2012.04733,
  title  = {CARAFE++: Unified Content-Aware ReAssembly of FEatures},
  author = {Jiaqi Wang and Kai Chen and Rui Xu and Ziwei Liu and Chen Change Loy and Dahua Lin},
  journal= {arXiv preprint arXiv:2012.04733},
  year   = {2020}
}

Comments

Technical Report. Extended journal version of the conference paper that appeared as arXiv:1905.02188