Research map / Detection and segmentation / CNN features
CARAFE++: Unified Content-Aware ReAssembly of FEatures
Jiaqi Wang, Kai Chen, Rui Xu, Ziwei Liu, Chen Change Loy, Dahua Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace strided convolution downsampling in ResNet-50 with CARAFE++ | Faster R-CNN ResNet-50 FPN · COCO val (1x) | AP box: 36.5 -> 38.8 | Improved |
| use CARAFE++ for downsampling instead of max pooling (ablation) | Faster R-CNN ResNet-50 FPN · COCO val (1x) | AP box: 37.8 -> 38.8 | Improved |
| replace nearest-neighbor upsampling in FPN with CARAFE++ (ablation) | Faster R-CNN ResNet-50 FPN · COCO val (1x) | AP box: 36.5 -> 37.8 | Improved |
| add an extra 3x3 conv after nearest or bilinear upsampling (ablation) | Faster R-CNN FPN upsampling · COCO val (1x) | AP box: 36.5 -> 36.6 | No clear effect |
| apply CARAFE++ in both backbone and FPN (ablation) | Faster R-CNN ResNet-50 · COCO val (1x) | AP box: 37.8 -> 39.6 | Improved |
| replace deconvolution in the mask head with CARAFE++ (ablation) | Mask R-CNN ResNet-50 · COCO val (1x) | AP mask: 34.2 -> 34.7 | Improved |
| apply CARAFE++ in FPN, mask head and backbone together (ablation) | Mask R-CNN ResNet-50 · COCO val (1x) | AP mask: 34.2 -> 36.8 | Improved |
| add CARAFE++ downsampling to ResNet backbone | ResNet-50 classification · ImageNet-1k val | Top-1 accuracy: 76.43 -> 77.46 | Improved |
| add CARAFE++ to ResNet-101 with DCNv2 detector | Faster R-CNN ResNet-101-DCNv2 · COCO test-dev (2x) | AP box: 43.2 -> 44.3 | Improved |
| add CARAFE++ to UperNet components PPM, FPN, FUSE and backbone | UperNet ResNet-50 · ADE20k val | mIoU: 40.44 -> 43.05 | Improved |
| add CARAFE++ backbone on top of PPM+FPN+FUSE (ablation) | UperNet ResNet-50 · ADE20k val | mIoU: 42.23 -> 43.05 | Improved |
| add CARAFE++ to Global&Local inpainting | Global&Local · Places val | PSNR: 19.58 -> 20.93 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CARAFE: Content-Aware ReAssembly of FEatures (ICCV 2019 · similarity 0.98)
- TAFE-Net: Task-Aware Feature Embeddings for Low Shot Learning (CVPR 2019 · similarity 0.85)
- Breathing New Life Into Small Object Detection With Detection-Oriented Rectification (TPAMI 2026 · similarity 0.84)
- FADE: A Task-Agnostic Upsampling Operator for Encoder–Decoder Architectures (IJCV 2024 · similarity 0.84)
- Deep Feature Pyramid Reconfiguration for Object Detection (ECCV 2018 · similarity 0.84)
- RDSNet: A New Deep Architecture forReciprocal Object Detection and Instance Segmentation (AAAI 2020 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
