Research map / Detection and segmentation / CNN features
Multi-Scale Location-Aware Kernel Representation 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 |
|---|---|---|---|
| Replace Faster R-CNN's first-order proposal representation with MLKP | Faster R-CNN with VGG-16 · PASCAL VOC 2007 | mAP: +4.9% | Helped |
| Replace Faster R-CNN's first-order proposal representation with MLKP | Faster R-CNN with VGG-16 · PASCAL VOC 2012 | mAP: +5.1% | Helped |
| Replace Faster R-CNN's first-order proposal representation with MLKP | Faster R-CNN with VGG-16 · MS COCO | AP (IOU=[0.5:0.05:0.95]): +5.0% | Helped |
| Use modified multi-scale features conv5_3/2 + conv4_3/2 instead of conv5_3 (ablation) | Faster R-CNN · PASCAL VOC 2007 | mAP: 73.2 -> 76.5 | Helped |
| Add conv3_3 features to conv5_3 + conv4_3 (ablation) | Faster R-CNN multi-scale feature map · PASCAL VOC 2007 | mAP: 76.2 -> 76.3; 11 -> 6 FPS | Mixed |
| Add order-2 location-aware kernel features (ablation) | Faster R-CNN with conv5_3 features · PASCAL VOC 2007 | mAP: 73.2 -> 76.4 | Helped |
| Increase polynomial order from 2 to 3 (ablation) | MLKP · PASCAL VOC 2007 | mAP: 77.7 -> 78.1 | Helped |
| Increase order-3 representation dimension from 4096 to 8192 (ablation) | MLKP with conv5_3/2 + conv4_3/2 · PASCAL VOC 2007 | mAP: 78.1 -> 77.7 | Hurt |
| Add the location-weight network (ablation) | MLKP with conv5_3/2 + conv4_3/2 · PASCAL VOC 2007 | mAP: +0.8% | Helped |
| Use order-3 representation with dimension 4096 (ablation) | MLKP · PASCAL VOC 2007 | mAP: 77.3%; +0.3% to +0.8% | Mixed |
| Use ResNet-101 instead of VGG-16 | MLKP · MS COCO | AP (IOU: 0.5:0.95): 26.9 -> 28.6 | Helped |
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)
- CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective (CVPR 2024 · similarity 0.87)
- Object Detection With Location-Aware Deformable Convolution and Backward Attention Filtering (CVPR 2019 · similarity 0.87)
- Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021 · similarity 0.87)
- LocNet: Improving Localization Accuracy for Object Detection (CVPR 2016 · 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.