Research map / Detection and segmentation / CNN features
Object Detection With Location-Aware Deformable Convolution and Backward Attention Filtering
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 |
|---|---|---|---|
| Add location-aware context embedding to Faster R-CNN (ablation) | VGG-16 detector · KITTI validation | AP: +4.8% AP | Helped |
| Add backward attention filtering to Faster R-CNN (ablation) | VGG-16 detector · KITTI validation | AP: 2.8% improvement in AP | Helped |
| Combine context embedding and backward attention filtering (ablation) | VGG-16 detector · KITTI validation | AP (%) car, moderate: 87.09 -> 91.74 | Helped |
| Combine context embedding and backward attention filtering (ablation) | VGG-16 detector · PASCAL VOC 2007 | mAP (%): 73.2 -> 76.1 | Helped |
| Set deformable-convolution dilation D1=2 and offset-estimation dilation D2=1 (ablation) | Location-aware deformable convolution · KITTI validation, moderate | AP (%): 87.35/75.19/74.25 -> 90.66/81.83/78.09 | Helped |
| Increase deformable-convolution dilation from 1 to 2 (ablation) | Deformable convolution · KITTI validation, moderate | AP (%): 86.70/74.89/73.96 -> 89.26/80.84/77.25 | Helped |
| Replace standard convolution with location-aware deformable convolution at D1=2, D2=1 (ablation) | Context feature extraction · KITTI validation, moderate | AP (%) car: 87.09 -> 90.66 | Helped |
| Use backward attention filtering instead of weak semantic segmentation attention (ablation) | VGG-16 detector · KITTI validation, moderate | AP (%): 86.83/74.59/73.55 -> 90.32/78.44/77.80 | Helped |
| Use backward attention filtering instead of residual attention (ablation) | VGG-16 detector · KITTI validation, moderate | AP (%): 88.92/77.28/76.76 -> 90.32/78.44/77.80 | Helped |
| Reduce the number of ROIs using backward attention filtering (ablation) | Faster R-CNN-based detector · KITTI validation | mean AP | No clear effect |
| Use a larger number of ROIs (ablation) | Faster R-CNN-based detector · KITTI validation | runtime | Hurt |
Closest papers by meaning
- Multi-Scale Location-Aware Kernel Representation for Object Detection (CVPR 2018 · similarity 0.87)
- Context R-CNN: Long Term Temporal Context for Per-Camera Object Detection (CVPR 2020 · similarity 0.86)
- DALDet: Depth-Aware Learning Based Object Detection for Autonomous Driving (AAAI 2024 · similarity 0.86)
- Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021 · similarity 0.86)
- Seeing without Looking: Contextual Rescoring of Object Detections for AP Maximization (CVPR 2020 · similarity 0.86)
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.85)
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.