Research map / Detection and segmentation / CNN features
Improving Convolutional Networks With Self-Calibrated Convolutions
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 ResNet-50's 3×3 convolutions with self-calibrated convolutions | ResNet-50 · ImageNet-1K | Top-1 accuracy: 76.4 -> 77.8 | Helped |
| Replace ResNet-101's 3×3 convolutions with self-calibrated convolutions | ResNet-101 · ImageNet-1K | Top-1 accuracy: 78.0 -> 78.9 | Helped |
| Use self-calibrated convolutions in ResNeXt-50 | ResNeXt-50 · ImageNet-1K | Top-1 accuracy: 77.4 -> 78.3 | Helped |
| Replace ResNeXt-101 convolutions with self-calibrated convolutions | ResNeXt-101 · ImageNet-1K | Top-1 accuracy: 78.5 -> 79.2 | Helped |
| Add self-calibrated convolutions to SE-ResNet-50 | SE-ResNet-50 · ImageNet-1K | Top-1 accuracy: 77.2 -> 78.2 | Helped |
| Replace ResNet-50-FPN with SCNet-50-FPN | Faster R-CNN · COCO minival | Box AP: 37.6 -> 40.8 | Helped |
| Replace ResNet-101-FPN with SCNet-101-FPN | Faster R-CNN · COCO minival | Box AP: 39.9 -> 42.0 | Helped |
| Replace ResNet-50-FPN with SCNet-50-FPN | Mask R-CNN · COCO minival | Mask AP: 35.0 -> 37.2 | Helped |
| Replace ResNet-101-FPN with SCNet-101-FPN | Mask R-CNN · COCO minival | Mask AP: 36.7 -> 38.4 | Helped |
| Replace ResNet-50 backbone with SCNet-50 | Keypoint detector · COCO val2017, 256 × 192 | OKS-based mAP: 70.6 -> 72.1 | Helped |
| Replace ResNet-50 backbone with SCNet-50 | Keypoint detector · COCO val2017, 384 × 288 | OKS-based mAP: 71.9 -> 74.4 | Helped |
| Set self-calibration down-sampling rate to 4 rather than 1 (ablation) | SCNet-50 · ImageNet-1K | Top-1 accuracy: 77.38% -> 77.81% | Helped |
Closest papers by meaning
- Involution: Inverting the Inherence of Convolution for Visual Recognition (CVPR 2021 · similarity 0.90)
- On the Integration of Self-Attention and Convolution (CVPR 2022 · similarity 0.89)
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.88)
- ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases (ICML 2021 · similarity 0.88)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.88)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.87)
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.