CARAFE: Content-Aware ReAssembly of FEatures

TL;DR

CARAFE is a novel, content-aware feature upsampling operator that improves dense prediction tasks by aggregating contextual information with adaptive kernels, offering significant performance gains with minimal computational cost.

Contribution

This paper introduces CARAFE, a universal, lightweight, and effective feature upsampling operator that adaptively handles content and aggregates large contextual information for better dense predictions.

Findings

Consistent performance improvements across object detection, segmentation, and inpainting tasks.

Substantial gains of 1.1-1.8% in accuracy or 1.1dB in inpainting quality.

Negligible additional computational overhead.

Abstract

Feature upsampling is a key operation in a number of modern convolutional network architectures, e.g. feature pyramids. Its design is critical for dense prediction tasks such as object detection and semantic/instance segmentation. In this work, we propose Content-Aware ReAssembly of FEatures (CARAFE), a universal, lightweight and highly effective operator to fulfill this goal. CARAFE has several appealing properties: (1) Large field of view. Unlike previous works (e.g. bilinear interpolation) that only exploit sub-pixel neighborhood, CARAFE can aggregate contextual information within a large receptive field. (2) Content-aware handling. Instead of using a fixed kernel for all samples (e.g. deconvolution), CARAFE enables instance-specific content-aware handling, which generates adaptive kernels on-the-fly. (3) Lightweight and fast to compute. 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 inpainting. CARAFE shows consistent and substantial gains across all the tasks (1.2%, 1.3%, 1.8%, 1.1db respectively) with negligible computational overhead. It has great potential to serve as a strong building block for future research. It has great potential to serve as a strong building block for future research. Code and models are available at https://github.com/open-mmlab/mmdetection.