Research map / Detection and segmentation / CNN features
Adaptive Feeding: Achieving Fast and Accurate Detections by Adaptively Combining Object Detectors
Hong-Yu Zhou, Bin-Bin Gao, Jianxin Wu
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 |
|---|---|---|---|
| use proposal-based AF features instead of raw VGG-16 image classification (ablation) | SSD300+SSD500 AF · VOC07 test | mAP: 72.5 -> 75.0 | Improved |
| use class histogram (20) instead of per-proposal class probabilities (20-prob) (ablation) | SSD300+SSD500 AF · VOC07 test | mAP: 74.4 -> 75.0 | Improved |
| add bounding box size (4s) features to class histogram and confidence (ablation) | SSD300+SSD500 AF · VOC07 test | mAP: 74.3 -> 75.0 | Improved |
| use box size and position (4s) rather than corner coordinates (4c) (ablation) | SSD300+SSD500 AF · VOC07 test | mAP: 74.5 -> 75.0 | Improved |
| raise number of proposals K from 10 to 25 (ablation) | SSD300+SSD500 AF · VOC07 test | mAP: 74.7 -> 75.0 | Improved |
| use K=25 instead of K=50 proposals (ablation) | SSD300+SSD500 AF · VOC07 test | mAP: 74.8 -> 75.0 | Improved |
| lower hard-image sampling weight from 5.13 to 3 (fast mode) (ablation) | SSD300+SSD500 AF · VOC07 test | mAP and FPS: 75.0 -> 74.4 mAP, 27 -> 33 FPS | Mixed |
| lower hard-image sampling weight from 8.43 to 5 (fast mode) (ablation) | SSD300+R-FCN AF · VOC07 test | mAP and FPS: 78.3 -> 76.9 mAP, 17 -> 24 FPS | Mixed |
| adaptively combine SSD300 with SSD500 (accurate mode) instead of SSD300 alone | 300-500-A · VOC12 test | mAP: 68.9 -> 71.1 | Improved |
| adaptively combine SSD300 with R-FCN (accurate mode) instead of SSD300 alone | 300-R-FCN-A · MS COCO test-dev | AP: 20.8 -> 27.0 | Improved |
| adaptively combine SSD300 with SSD500 instead of simple ensemble | SSD300+SSD500 AF · VOC07 test | mAP: 73.0 -> 75.0 | Improved |
| accurate-mode AF with R-FCN partner instead of R-FCN alone (speed-up at accuracy cost) | 300-R-FCN-A · VOC07 test | mAP and FPS: 79.0 -> 78.3 mAP, 8 -> 17 FPS | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- R-FCN — R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016) · better on 0 of 3 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
- Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors (CVPR 2017 · similarity 0.87)
- Training-Time-Friendly Network for Real-Time Object Detection (AAAI 2020 · similarity 0.87)
- Test-Time Adaptive Object Detection with Foundation Model (NeurIPS 2025 · similarity 0.85)
- Plug and Play Active Learning for Object Detection (CVPR 2024 · similarity 0.85)
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection (NeurIPS 2024 · similarity 0.84)
- LEOD: Label-Efficient Object Detection for Event Cameras (CVPR 2024 · 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.
