Research map / Detection and segmentation / CNN features
Feature Selective Networks for Object Detection
Read the paperCheck 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 |
|---|---|---|---|
| Add sub-region attention (ablation) | Feature Selective Network · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 78.8 -> 82.2 | Helped |
| Add aspect ratio attention (ablation) | Feature Selective Network · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 78.8 -> 81.0 | Helped |
| Combine sub-region and aspect ratio attention (ablation) | Feature Selective Network · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.2 -> 82.9 | Helped |
| Add sub-region and aspect ratio attention (ablation) | Faster R-CNN · PASCAL VOC 2007 test, GoogLeNet | mAP (%): 74.8 -> 76.8 | Helped |
| Add sub-region and aspect ratio attention (ablation) | Faster R-CNN · PASCAL VOC 2007 test, VGG-16 | mAP (%): 73.2 -> 74.3 | Helped |
| Add dimension reduction to compact RoI features (ablation) | Faster R-CNN with simplified detection subnetwork · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 78.8 -> 79.5 | Helped |
| Set selective channel number to 20 rather than 40 (ablation) | Feature Selective Network · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.9 -> 82.5 | No clear effect |
| Reduce selective channel number to 5 (ablation) | Feature Selective Network · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.9 -> 80.5 | Hurt |
| Reduce selective channel number to 1 (ablation) | Feature Selective Network · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.9 -> 79.4 | Hurt |
| Use a 3x3 standard convolution instead of 1x1 convolution (ablation) | Sub-region attention bank generation · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.3 -> 82.6 | Helped |
| Use shifted 3x3 convolution with shift toward the assumed RoI center (ablation) | Sub-region attention bank generation · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.6 -> 82.9 | Helped |
| Shift convolution outward from the RoI center (ablation) | Sub-region attention bank generation · PASCAL VOC 2007 test, ResNet-101 | mAP (%): 82.9 -> 82.5 | Hurt |
Closest papers by meaning
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.88)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · similarity 0.88)
- AugFPN: Improving Multi-Scale Feature Learning for Object Detection (CVPR 2020 · similarity 0.87)
- M2Det: A Single-Shot Object Detector Based on Multi-Level Feature Pyramid Network (AAAI 2019 · similarity 0.87)
- RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder (NeurIPS 2020 · similarity 0.87)
- Semantic-Aware Transformation-Invariant RoI Align (AAAI 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.