# CARAFE: Content-Aware ReAssembly of FEatures

arXiv: https://arxiv.org/abs/1905.02188 · HTML: https://mlsanity.com/paper/1905.02188/

## Key facts

- Cited by 119 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 6 May 2019.

## Built on by (most cited first, 30 of 119)

- [Dynamic Neural Networks: A Survey](https://mlsanity.com/paper/2102.04906/) (2021)
- [Towards Large-Scale Small Object Detection: Survey and Benchmarks](https://mlsanity.com/paper/2207.14096/) (2022)
- [FeatUp: A Model-Agnostic Framework for Features at Any Resolution](https://mlsanity.com/paper/2403.10516/) (2024)
- [Involution: Inverting the Inherence of Convolution for Visual Recognition](https://mlsanity.com/paper/2103.06255/) (2021)
- [MonoRUn: Monocular 3D Object Detection by Reconstruction and Uncertainty Propagation](https://mlsanity.com/paper/2103.12605/) (2021)
- [V3Det: Vast Vocabulary Visual Detection Dataset](https://mlsanity.com/paper/2304.03752/) (2023)
- [Learning to Upsample by Learning to Sample](https://mlsanity.com/paper/2308.15085/) (2023)
- [Decoupled Dynamic Filter Networks](https://mlsanity.com/paper/2104.14107/) (2021)
- [Frequency-aware Feature Fusion for Dense Image Prediction](https://mlsanity.com/paper/2408.12879/) (2024)
- [Co-advise: Cross Inductive Bias Distillation](https://mlsanity.com/paper/2106.12378/) (2021)
- [DUET: A Tuning-Free Device-Cloud Collaborative Parameters Generation Framework for Efficient Device Model Generalization](https://mlsanity.com/paper/2209.05227/) (2022)
- [FSGANv2: Improved Subject Agnostic Face Swapping and Reenactment](https://mlsanity.com/paper/2202.12972/) (2022)
- [Few-shot Medical Image Segmentation with Cycle-resemblance Attention](https://mlsanity.com/paper/2212.03967/) (2022)
- [KBNet: Kernel Basis Network for Image Restoration](https://mlsanity.com/paper/2303.02881/) (2023)
- [SAPA: Similarity-Aware Point Affiliation for Feature Upsampling](https://mlsanity.com/paper/2209.12866/) (2022)
- [Differentiable Multi-Granularity Human Representation Learning for Instance-Aware Human Semantic Parsing](https://mlsanity.com/paper/2103.04570/) (2021)
- [Removing Diffraction Image Artifacts in Under-Display Camera via Dynamic Skip Connection Network](https://mlsanity.com/paper/2104.09556/) (2021)
- [Learning Implicit Feature Alignment Function for Semantic Segmentation](https://mlsanity.com/paper/2206.08655/) (2022)
- [PointConvFormer: Revenge of the Point-based Convolution](https://mlsanity.com/paper/2208.02879/) (2022)
- [AnyUp: Universal Feature Upsampling](https://mlsanity.com/paper/2510.12764/) (2025)
- [LiFT: A Surprisingly Simple Lightweight Feature Transform for Dense ViT Descriptors](https://mlsanity.com/paper/2403.14625/) (2024)
- [More than Encoder: Introducing Transformer Decoder to Upsample](https://mlsanity.com/paper/2106.10637/) (2021)
- [A Decomposition Model for Stereo Matching](https://mlsanity.com/paper/2104.07516/) (2021)
- [FADE: Fusing the Assets of Decoder and Encoder for Task-Agnostic Upsampling](https://mlsanity.com/paper/2207.10392/) (2022)
- [LoftUp: Learning a Coordinate-Based Feature Upsampler for Vision Foundation Models](https://mlsanity.com/paper/2504.14032/) (2025)
- [Rank & Sort Loss for Object Detection and Instance Segmentation](https://mlsanity.com/paper/2107.11669/) (2021)
- [OmniCity: Omnipotent City Understanding with Multi-level and Multi-view Images](https://mlsanity.com/paper/2208.00928/) (2022)
- [JAFAR: Jack up Any Feature at Any Resolution](https://mlsanity.com/paper/2506.11136/) (2025)
- [Frequency Dynamic Convolution for Dense Image Prediction](https://mlsanity.com/paper/2503.18783/) (2025)
- [ELSA: Enhanced Local Self-Attention for Vision Transformer](https://mlsanity.com/paper/2112.12786/) (2021)

## Abstract (arXiv)

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.